git.oblomov.eu Git - linux-2.6/blob - drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/config.h>
  37 #include <linux/kthread.h>
  38 #include <linux/linkage.h>
  39 #include <linux/raid/md.h>
  40 #include <linux/raid/bitmap.h>
  41 #include <linux/sysctl.h>
  42 #include <linux/devfs_fs_kernel.h>
  43 #include <linux/buffer_head.h> /* for invalidate_bdev */
  44 #include <linux/suspend.h>
  45
  46 #include <linux/init.h>
  47
  48 #include <linux/file.h>
  49
  50 #ifdef CONFIG_KMOD
  51 #include <linux/kmod.h>
  52 #endif
  53
  54 #include <asm/unaligned.h>
  55
  56 #define MAJOR_NR MD_MAJOR
  57 #define MD_DRIVER
  58
  59 /* 63 partitions with the alternate major number (mdp) */
  60 #define MdpMinorShift 6
  61
  62 #define DEBUG 0
  63 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  64
  65
  66 #ifndef MODULE
  67 static void autostart_arrays (int part);
  68 #endif
  69
  70 static mdk_personality_t *pers[MAX_PERSONALITY];
  71 static DEFINE_SPINLOCK(pers_lock);
  72
  73 /*
  74  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  75  * is 1000 KB/sec, so the extra system load does not show up that much.
  76  * Increase it if you want to have more _guaranteed_ speed. Note that
  77  * the RAID driver will use the maximum available bandwith if the IO
  78  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  79  * speed limit - in case reconstruction slows down your system despite
  80  * idle IO detection.
  81  *
  82  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  83  */
  84
  85 static int sysctl_speed_limit_min = 1000;
  86 static int sysctl_speed_limit_max = 200000;
  87
  88 static struct ctl_table_header *raid_table_header;
  89
  90 static ctl_table raid_table[] = {
  91         {
  92                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
  93                 .procname       = "speed_limit_min",
  94                 .data           = &sysctl_speed_limit_min,
  95                 .maxlen         = sizeof(int),
  96                 .mode           = 0644,
  97                 .proc_handler   = &proc_dointvec,
  98         },
  99         {
 100                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 101                 .procname       = "speed_limit_max",
 102                 .data           = &sysctl_speed_limit_max,
 103                 .maxlen         = sizeof(int),
 104                 .mode           = 0644,
 105                 .proc_handler   = &proc_dointvec,
 106         },
 107         { .ctl_name = 0 }
 108 };
 109
 110 static ctl_table raid_dir_table[] = {
 111         {
 112                 .ctl_name       = DEV_RAID,
 113                 .procname       = "raid",
 114                 .maxlen         = 0,
 115                 .mode           = 0555,
 116                 .child          = raid_table,
 117         },
 118         { .ctl_name = 0 }
 119 };
 120
 121 static ctl_table raid_root_table[] = {
 122         {
 123                 .ctl_name       = CTL_DEV,
 124                 .procname       = "dev",
 125                 .maxlen         = 0,
 126                 .mode           = 0555,
 127                 .child          = raid_dir_table,
 128         },
 129         { .ctl_name = 0 }
 130 };
 131
 132 static struct block_device_operations md_fops;
 133
 134 /*
 135  * Enables to iterate over all existing md arrays
 136  * all_mddevs_lock protects this list.
 137  */
 138 static LIST_HEAD(all_mddevs);
 139 static DEFINE_SPINLOCK(all_mddevs_lock);
 140
 141
 142 /*
 143  * iterates through all used mddevs in the system.
 144  * We take care to grab the all_mddevs_lock whenever navigating
 145  * the list, and to always hold a refcount when unlocked.
 146  * Any code which breaks out of this loop while own
 147  * a reference to the current mddev and must mddev_put it.
 148  */
 149 #define ITERATE_MDDEV(mddev,tmp)                                        \
 150                                                                         \
 151         for (({ spin_lock(&all_mddevs_lock);                            \
 152                 tmp = all_mddevs.next;                                  \
 153                 mddev = NULL;});                                        \
 154              ({ if (tmp != &all_mddevs)                                 \
 155                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 156                 spin_unlock(&all_mddevs_lock);                          \
 157                 if (mddev) mddev_put(mddev);                            \
 158                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 159                 tmp != &all_mddevs;});                                  \
 160              ({ spin_lock(&all_mddevs_lock);                            \
 161                 tmp = tmp->next;})                                      \
 162                 )
 163
 164
 165 static int md_fail_request (request_queue_t *q, struct bio *bio)
 166 {
 167         bio_io_error(bio, bio->bi_size);
 168         return 0;
 169 }
 170
 171 static inline mddev_t *mddev_get(mddev_t *mddev)
 172 {
 173         atomic_inc(&mddev->active);
 174         return mddev;
 175 }
 176
 177 static void mddev_put(mddev_t *mddev)
 178 {
 179         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 180                 return;
 181         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 182                 list_del(&mddev->all_mddevs);
 183                 blk_put_queue(mddev->queue);
 184                 kfree(mddev);
 185         }
 186         spin_unlock(&all_mddevs_lock);
 187 }
 188
 189 static mddev_t * mddev_find(dev_t unit)
 190 {
 191         mddev_t *mddev, *new = NULL;
 192
 193  retry:
 194         spin_lock(&all_mddevs_lock);
 195         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 196                 if (mddev->unit == unit) {
 197                         mddev_get(mddev);
 198                         spin_unlock(&all_mddevs_lock);
 199                         kfree(new);
 200                         return mddev;
 201                 }
 202
 203         if (new) {
 204                 list_add(&new->all_mddevs, &all_mddevs);
 205                 spin_unlock(&all_mddevs_lock);
 206                 return new;
 207         }
 208         spin_unlock(&all_mddevs_lock);
 209
 210         new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
 211         if (!new)
 212                 return NULL;
 213
 214         memset(new, 0, sizeof(*new));
 215
 216         new->unit = unit;
 217         if (MAJOR(unit) == MD_MAJOR)
 218                 new->md_minor = MINOR(unit);
 219         else
 220                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 221
 222         init_MUTEX(&new->reconfig_sem);
 223         INIT_LIST_HEAD(&new->disks);
 224         INIT_LIST_HEAD(&new->all_mddevs);
 225         init_timer(&new->safemode_timer);
 226         atomic_set(&new->active, 1);
 227         spin_lock_init(&new->write_lock);
 228         init_waitqueue_head(&new->sb_wait);
 229
 230         new->queue = blk_alloc_queue(GFP_KERNEL);
 231         if (!new->queue) {
 232                 kfree(new);
 233                 return NULL;
 234         }
 235
 236         blk_queue_make_request(new->queue, md_fail_request);
 237
 238         goto retry;
 239 }
 240
 241 static inline int mddev_lock(mddev_t * mddev)
 242 {
 243         return down_interruptible(&mddev->reconfig_sem);
 244 }
 245
 246 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 247 {
 248         down(&mddev->reconfig_sem);
 249 }
 250
 251 static inline int mddev_trylock(mddev_t * mddev)
 252 {
 253         return down_trylock(&mddev->reconfig_sem);
 254 }
 255
 256 static inline void mddev_unlock(mddev_t * mddev)
 257 {
 258         up(&mddev->reconfig_sem);
 259
 260         md_wakeup_thread(mddev->thread);
 261 }
 262
 263 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 264 {
 265         mdk_rdev_t * rdev;
 266         struct list_head *tmp;
 267
 268         ITERATE_RDEV(mddev,rdev,tmp) {
 269                 if (rdev->desc_nr == nr)
 270                         return rdev;
 271         }
 272         return NULL;
 273 }
 274
 275 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 276 {
 277         struct list_head *tmp;
 278         mdk_rdev_t *rdev;
 279
 280         ITERATE_RDEV(mddev,rdev,tmp) {
 281                 if (rdev->bdev->bd_dev == dev)
 282                         return rdev;
 283         }
 284         return NULL;
 285 }
 286
 287 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 288 {
 289         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 290         return MD_NEW_SIZE_BLOCKS(size);
 291 }
 292
 293 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 294 {
 295         sector_t size;
 296
 297         size = rdev->sb_offset;
 298
 299         if (chunk_size)
 300                 size &= ~((sector_t)chunk_size/1024 - 1);
 301         return size;
 302 }
 303
 304 static int alloc_disk_sb(mdk_rdev_t * rdev)
 305 {
 306         if (rdev->sb_page)
 307                 MD_BUG();
 308
 309         rdev->sb_page = alloc_page(GFP_KERNEL);
 310         if (!rdev->sb_page) {
 311                 printk(KERN_ALERT "md: out of memory.\n");
 312                 return -EINVAL;
 313         }
 314
 315         return 0;
 316 }
 317
 318 static void free_disk_sb(mdk_rdev_t * rdev)
 319 {
 320         if (rdev->sb_page) {
 321                 page_cache_release(rdev->sb_page);
 322                 rdev->sb_loaded = 0;
 323                 rdev->sb_page = NULL;
 324                 rdev->sb_offset = 0;
 325                 rdev->size = 0;
 326         }
 327 }
 328
 329
 330 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 331 {
 332         mdk_rdev_t *rdev = bio->bi_private;
 333         if (bio->bi_size)
 334                 return 1;
 335
 336         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
 337                 md_error(rdev->mddev, rdev);
 338
 339         if (atomic_dec_and_test(&rdev->mddev->pending_writes))
 340                 wake_up(&rdev->mddev->sb_wait);
 341         bio_put(bio);
 342         return 0;
 343 }
 344
 345 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 346                    sector_t sector, int size, struct page *page)
 347 {
 348         /* write first size bytes of page to sector of rdev
 349          * Increment mddev->pending_writes before returning
 350          * and decrement it on completion, waking up sb_wait
 351          * if zero is reached.
 352          * If an error occurred, call md_error
 353          */
 354         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 355
 356         bio->bi_bdev = rdev->bdev;
 357         bio->bi_sector = sector;
 358         bio_add_page(bio, page, size, 0);
 359         bio->bi_private = rdev;
 360         bio->bi_end_io = super_written;
 361         atomic_inc(&mddev->pending_writes);
 362         submit_bio((1<<BIO_RW)|(1<<BIO_RW_SYNC), bio);
 363 }
 364
 365 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 366 {
 367         if (bio->bi_size)
 368                 return 1;
 369
 370         complete((struct completion*)bio->bi_private);
 371         return 0;
 372 }
 373
 374 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 375                    struct page *page, int rw)
 376 {
 377         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 378         struct completion event;
 379         int ret;
 380
 381         rw |= (1 << BIO_RW_SYNC);
 382
 383         bio->bi_bdev = bdev;
 384         bio->bi_sector = sector;
 385         bio_add_page(bio, page, size, 0);
 386         init_completion(&event);
 387         bio->bi_private = &event;
 388         bio->bi_end_io = bi_complete;
 389         submit_bio(rw, bio);
 390         wait_for_completion(&event);
 391
 392         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 393         bio_put(bio);
 394         return ret;
 395 }
 396
 397 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 398 {
 399         char b[BDEVNAME_SIZE];
 400         if (!rdev->sb_page) {
 401                 MD_BUG();
 402                 return -EINVAL;
 403         }
 404         if (rdev->sb_loaded)
 405                 return 0;
 406
 407
 408         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 409                 goto fail;
 410         rdev->sb_loaded = 1;
 411         return 0;
 412
 413 fail:
 414         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 415                 bdevname(rdev->bdev,b));
 416         return -EINVAL;
 417 }
 418
 419 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 420 {
 421         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 422                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 423                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 424                 (sb1->set_uuid3 == sb2->set_uuid3))
 425
 426                 return 1;
 427
 428         return 0;
 429 }
 430
 431
 432 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 433 {
 434         int ret;
 435         mdp_super_t *tmp1, *tmp2;
 436
 437         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 438         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 439
 440         if (!tmp1 || !tmp2) {
 441                 ret = 0;
 442                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 443                 goto abort;
 444         }
 445
 446         *tmp1 = *sb1;
 447         *tmp2 = *sb2;
 448
 449         /*
 450          * nr_disks is not constant
 451          */
 452         tmp1->nr_disks = 0;
 453         tmp2->nr_disks = 0;
 454
 455         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 456                 ret = 0;
 457         else
 458                 ret = 1;
 459
 460 abort:
 461         kfree(tmp1);
 462         kfree(tmp2);
 463         return ret;
 464 }
 465
 466 static unsigned int calc_sb_csum(mdp_super_t * sb)
 467 {
 468         unsigned int disk_csum, csum;
 469
 470         disk_csum = sb->sb_csum;
 471         sb->sb_csum = 0;
 472         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 473         sb->sb_csum = disk_csum;
 474         return csum;
 475 }
 476
 477
 478 /*
 479  * Handle superblock details.
 480  * We want to be able to handle multiple superblock formats
 481  * so we have a common interface to them all, and an array of
 482  * different handlers.
 483  * We rely on user-space to write the initial superblock, and support
 484  * reading and updating of superblocks.
 485  * Interface methods are:
 486  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 487  *      loads and validates a superblock on dev.
 488  *      if refdev != NULL, compare superblocks on both devices
 489  *    Return:
 490  *      0 - dev has a superblock that is compatible with refdev
 491  *      1 - dev has a superblock that is compatible and newer than refdev
 492  *          so dev should be used as the refdev in future
 493  *     -EINVAL superblock incompatible or invalid
 494  *     -othererror e.g. -EIO
 495  *
 496  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 497  *      Verify that dev is acceptable into mddev.
 498  *       The first time, mddev->raid_disks will be 0, and data from
 499  *       dev should be merged in.  Subsequent calls check that dev
 500  *       is new enough.  Return 0 or -EINVAL
 501  *
 502  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 503  *     Update the superblock for rdev with data in mddev
 504  *     This does not write to disc.
 505  *
 506  */
 507
 508 struct super_type  {
 509         char            *name;
 510         struct module   *owner;
 511         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 512         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 513         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 514 };
 515
 516 /*
 517  * load_super for 0.90.0
 518  */
 519 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 520 {
 521         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 522         mdp_super_t *sb;
 523         int ret;
 524         sector_t sb_offset;
 525
 526         /*
 527          * Calculate the position of the superblock,
 528          * it's at the end of the disk.
 529          *
 530          * It also happens to be a multiple of 4Kb.
 531          */
 532         sb_offset = calc_dev_sboffset(rdev->bdev);
 533         rdev->sb_offset = sb_offset;
 534
 535         ret = read_disk_sb(rdev, MD_SB_BYTES);
 536         if (ret) return ret;
 537
 538         ret = -EINVAL;
 539
 540         bdevname(rdev->bdev, b);
 541         sb = (mdp_super_t*)page_address(rdev->sb_page);
 542
 543         if (sb->md_magic != MD_SB_MAGIC) {
 544                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 545                        b);
 546                 goto abort;
 547         }
 548
 549         if (sb->major_version != 0 ||
 550             sb->minor_version != 90) {
 551                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 552                         sb->major_version, sb->minor_version,
 553                         b);
 554                 goto abort;
 555         }
 556
 557         if (sb->raid_disks <= 0)
 558                 goto abort;
 559
 560         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 561                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 562                         b);
 563                 goto abort;
 564         }
 565
 566         rdev->preferred_minor = sb->md_minor;
 567         rdev->data_offset = 0;
 568         rdev->sb_size = MD_SB_BYTES;
 569
 570         if (sb->level == LEVEL_MULTIPATH)
 571                 rdev->desc_nr = -1;
 572         else
 573                 rdev->desc_nr = sb->this_disk.number;
 574
 575         if (refdev == 0)
 576                 ret = 1;
 577         else {
 578                 __u64 ev1, ev2;
 579                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 580                 if (!uuid_equal(refsb, sb)) {
 581                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 582                                 b, bdevname(refdev->bdev,b2));
 583                         goto abort;
 584                 }
 585                 if (!sb_equal(refsb, sb)) {
 586                         printk(KERN_WARNING "md: %s has same UUID"
 587                                " but different superblock to %s\n",
 588                                b, bdevname(refdev->bdev, b2));
 589                         goto abort;
 590                 }
 591                 ev1 = md_event(sb);
 592                 ev2 = md_event(refsb);
 593                 if (ev1 > ev2)
 594                         ret = 1;
 595                 else
 596                         ret = 0;
 597         }
 598         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 599
 600  abort:
 601         return ret;
 602 }
 603
 604 /*
 605  * validate_super for 0.90.0
 606  */
 607 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 608 {
 609         mdp_disk_t *desc;
 610         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 611
 612         rdev->raid_disk = -1;
 613         rdev->in_sync = 0;
 614         if (mddev->raid_disks == 0) {
 615                 mddev->major_version = 0;
 616                 mddev->minor_version = sb->minor_version;
 617                 mddev->patch_version = sb->patch_version;
 618                 mddev->persistent = ! sb->not_persistent;
 619                 mddev->chunk_size = sb->chunk_size;
 620                 mddev->ctime = sb->ctime;
 621                 mddev->utime = sb->utime;
 622                 mddev->level = sb->level;
 623                 mddev->layout = sb->layout;
 624                 mddev->raid_disks = sb->raid_disks;
 625                 mddev->size = sb->size;
 626                 mddev->events = md_event(sb);
 627                 mddev->bitmap_offset = 0;
 628                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 629
 630                 if (sb->state & (1<<MD_SB_CLEAN))
 631                         mddev->recovery_cp = MaxSector;
 632                 else {
 633                         if (sb->events_hi == sb->cp_events_hi &&
 634                                 sb->events_lo == sb->cp_events_lo) {
 635                                 mddev->recovery_cp = sb->recovery_cp;
 636                         } else
 637                                 mddev->recovery_cp = 0;
 638                 }
 639
 640                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 641                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 642                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 643                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 644
 645                 mddev->max_disks = MD_SB_DISKS;
 646
 647                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 648                     mddev->bitmap_file == NULL) {
 649                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6) {
 650                                 /* FIXME use a better test */
 651                                 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
 652                                 return -EINVAL;
 653                         }
 654                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 655                 }
 656
 657         } else if (mddev->pers == NULL) {
 658                 /* Insist on good event counter while assembling */
 659                 __u64 ev1 = md_event(sb);
 660                 ++ev1;
 661                 if (ev1 < mddev->events)
 662                         return -EINVAL;
 663         } else if (mddev->bitmap) {
 664                 /* if adding to array with a bitmap, then we can accept an
 665                  * older device ... but not too old.
 666                  */
 667                 __u64 ev1 = md_event(sb);
 668                 if (ev1 < mddev->bitmap->events_cleared)
 669                         return 0;
 670         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 671                 return 0;
 672
 673         if (mddev->level != LEVEL_MULTIPATH) {
 674                 rdev->faulty = 0;
 675                 rdev->flags = 0;
 676                 desc = sb->disks + rdev->desc_nr;
 677
 678                 if (desc->state & (1<<MD_DISK_FAULTY))
 679                         rdev->faulty = 1;
 680                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 681                          desc->raid_disk < mddev->raid_disks) {
 682                         rdev->in_sync = 1;
 683                         rdev->raid_disk = desc->raid_disk;
 684                 }
 685                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 686                         set_bit(WriteMostly, &rdev->flags);
 687         } else /* MULTIPATH are always insync */
 688                 rdev->in_sync = 1;
 689         return 0;
 690 }
 691
 692 /*
 693  * sync_super for 0.90.0
 694  */
 695 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 696 {
 697         mdp_super_t *sb;
 698         struct list_head *tmp;
 699         mdk_rdev_t *rdev2;
 700         int next_spare = mddev->raid_disks;
 701
 702         /* make rdev->sb match mddev data..
 703          *
 704          * 1/ zero out disks
 705          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 706          * 3/ any empty disks < next_spare become removed
 707          *
 708          * disks[0] gets initialised to REMOVED because
 709          * we cannot be sure from other fields if it has
 710          * been initialised or not.
 711          */
 712         int i;
 713         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 714
 715         sb = (mdp_super_t*)page_address(rdev->sb_page);
 716
 717         memset(sb, 0, sizeof(*sb));
 718
 719         sb->md_magic = MD_SB_MAGIC;
 720         sb->major_version = mddev->major_version;
 721         sb->minor_version = mddev->minor_version;
 722         sb->patch_version = mddev->patch_version;
 723         sb->gvalid_words  = 0; /* ignored */
 724         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 725         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 726         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 727         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 728
 729         sb->ctime = mddev->ctime;
 730         sb->level = mddev->level;
 731         sb->size  = mddev->size;
 732         sb->raid_disks = mddev->raid_disks;
 733         sb->md_minor = mddev->md_minor;
 734         sb->not_persistent = !mddev->persistent;
 735         sb->utime = mddev->utime;
 736         sb->state = 0;
 737         sb->events_hi = (mddev->events>>32);
 738         sb->events_lo = (u32)mddev->events;
 739
 740         if (mddev->in_sync)
 741         {
 742                 sb->recovery_cp = mddev->recovery_cp;
 743                 sb->cp_events_hi = (mddev->events>>32);
 744                 sb->cp_events_lo = (u32)mddev->events;
 745                 if (mddev->recovery_cp == MaxSector)
 746                         sb->state = (1<< MD_SB_CLEAN);
 747         } else
 748                 sb->recovery_cp = 0;
 749
 750         sb->layout = mddev->layout;
 751         sb->chunk_size = mddev->chunk_size;
 752
 753         if (mddev->bitmap && mddev->bitmap_file == NULL)
 754                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 755
 756         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 757         ITERATE_RDEV(mddev,rdev2,tmp) {
 758                 mdp_disk_t *d;
 759                 if (rdev2->raid_disk >= 0 && rdev2->in_sync && !rdev2->faulty)
 760                         rdev2->desc_nr = rdev2->raid_disk;
 761                 else
 762                         rdev2->desc_nr = next_spare++;
 763                 d = &sb->disks[rdev2->desc_nr];
 764                 nr_disks++;
 765                 d->number = rdev2->desc_nr;
 766                 d->major = MAJOR(rdev2->bdev->bd_dev);
 767                 d->minor = MINOR(rdev2->bdev->bd_dev);
 768                 if (rdev2->raid_disk >= 0 && rdev->in_sync && !rdev2->faulty)
 769                         d->raid_disk = rdev2->raid_disk;
 770                 else
 771                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 772                 if (rdev2->faulty) {
 773                         d->state = (1<<MD_DISK_FAULTY);
 774                         failed++;
 775                 } else if (rdev2->in_sync) {
 776                         d->state = (1<<MD_DISK_ACTIVE);
 777                         d->state |= (1<<MD_DISK_SYNC);
 778                         active++;
 779                         working++;
 780                 } else {
 781                         d->state = 0;
 782                         spare++;
 783                         working++;
 784                 }
 785                 if (test_bit(WriteMostly, &rdev2->flags))
 786                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 787         }
 788
 789         /* now set the "removed" and "faulty" bits on any missing devices */
 790         for (i=0 ; i < mddev->raid_disks ; i++) {
 791                 mdp_disk_t *d = &sb->disks[i];
 792                 if (d->state == 0 && d->number == 0) {
 793                         d->number = i;
 794                         d->raid_disk = i;
 795                         d->state = (1<<MD_DISK_REMOVED);
 796                         d->state |= (1<<MD_DISK_FAULTY);
 797                         failed++;
 798                 }
 799         }
 800         sb->nr_disks = nr_disks;
 801         sb->active_disks = active;
 802         sb->working_disks = working;
 803         sb->failed_disks = failed;
 804         sb->spare_disks = spare;
 805
 806         sb->this_disk = sb->disks[rdev->desc_nr];
 807         sb->sb_csum = calc_sb_csum(sb);
 808 }
 809
 810 /*
 811  * version 1 superblock
 812  */
 813
 814 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 815 {
 816         unsigned int disk_csum, csum;
 817         unsigned long long newcsum;
 818         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 819         unsigned int *isuper = (unsigned int*)sb;
 820         int i;
 821
 822         disk_csum = sb->sb_csum;
 823         sb->sb_csum = 0;
 824         newcsum = 0;
 825         for (i=0; size>=4; size -= 4 )
 826                 newcsum += le32_to_cpu(*isuper++);
 827
 828         if (size == 2)
 829                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 830
 831         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 832         sb->sb_csum = disk_csum;
 833         return cpu_to_le32(csum);
 834 }
 835
 836 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 837 {
 838         struct mdp_superblock_1 *sb;
 839         int ret;
 840         sector_t sb_offset;
 841         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 842         int bmask;
 843
 844         /*
 845          * Calculate the position of the superblock.
 846          * It is always aligned to a 4K boundary and
 847          * depeding on minor_version, it can be:
 848          * 0: At least 8K, but less than 12K, from end of device
 849          * 1: At start of device
 850          * 2: 4K from start of device.
 851          */
 852         switch(minor_version) {
 853         case 0:
 854                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
 855                 sb_offset -= 8*2;
 856                 sb_offset &= ~(sector_t)(4*2-1);
 857                 /* convert from sectors to K */
 858                 sb_offset /= 2;
 859                 break;
 860         case 1:
 861                 sb_offset = 0;
 862                 break;
 863         case 2:
 864                 sb_offset = 4;
 865                 break;
 866         default:
 867                 return -EINVAL;
 868         }
 869         rdev->sb_offset = sb_offset;
 870
 871         /* superblock is rarely larger than 1K, but it can be larger,
 872          * and it is safe to read 4k, so we do that
 873          */
 874         ret = read_disk_sb(rdev, 4096);
 875         if (ret) return ret;
 876
 877
 878         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 879
 880         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
 881             sb->major_version != cpu_to_le32(1) ||
 882             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
 883             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
 884             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
 885                 return -EINVAL;
 886
 887         if (calc_sb_1_csum(sb) != sb->sb_csum) {
 888                 printk("md: invalid superblock checksum on %s\n",
 889                         bdevname(rdev->bdev,b));
 890                 return -EINVAL;
 891         }
 892         if (le64_to_cpu(sb->data_size) < 10) {
 893                 printk("md: data_size too small on %s\n",
 894                        bdevname(rdev->bdev,b));
 895                 return -EINVAL;
 896         }
 897         rdev->preferred_minor = 0xffff;
 898         rdev->data_offset = le64_to_cpu(sb->data_offset);
 899
 900         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
 901         bmask = block_size(rdev->bdev)-1;
 902         if (rdev->sb_size & bmask)
 903                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
 904
 905         if (refdev == 0)
 906                 return 1;
 907         else {
 908                 __u64 ev1, ev2;
 909                 struct mdp_superblock_1 *refsb =
 910                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
 911
 912                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
 913                     sb->level != refsb->level ||
 914                     sb->layout != refsb->layout ||
 915                     sb->chunksize != refsb->chunksize) {
 916                         printk(KERN_WARNING "md: %s has strangely different"
 917                                 " superblock to %s\n",
 918                                 bdevname(rdev->bdev,b),
 919                                 bdevname(refdev->bdev,b2));
 920                         return -EINVAL;
 921                 }
 922                 ev1 = le64_to_cpu(sb->events);
 923                 ev2 = le64_to_cpu(refsb->events);
 924
 925                 if (ev1 > ev2)
 926                         return 1;
 927         }
 928         if (minor_version)
 929                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
 930         else
 931                 rdev->size = rdev->sb_offset;
 932         if (rdev->size < le64_to_cpu(sb->data_size)/2)
 933                 return -EINVAL;
 934         rdev->size = le64_to_cpu(sb->data_size)/2;
 935         if (le32_to_cpu(sb->chunksize))
 936                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
 937         return 0;
 938 }
 939
 940 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 941 {
 942         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
 943
 944         rdev->raid_disk = -1;
 945         rdev->in_sync = 0;
 946         if (mddev->raid_disks == 0) {
 947                 mddev->major_version = 1;
 948                 mddev->patch_version = 0;
 949                 mddev->persistent = 1;
 950                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
 951                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
 952                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
 953                 mddev->level = le32_to_cpu(sb->level);
 954                 mddev->layout = le32_to_cpu(sb->layout);
 955                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
 956                 mddev->size = le64_to_cpu(sb->size)/2;
 957                 mddev->events = le64_to_cpu(sb->events);
 958                 mddev->bitmap_offset = 0;
 959                 mddev->default_bitmap_offset = 0;
 960                 if (mddev->minor_version == 0)
 961                         mddev->default_bitmap_offset = -(64*1024)/512;
 962
 963                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
 964                 memcpy(mddev->uuid, sb->set_uuid, 16);
 965
 966                 mddev->max_disks =  (4096-256)/2;
 967
 968                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
 969                     mddev->bitmap_file == NULL ) {
 970                         if (mddev->level != 1) {
 971                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
 972                                 return -EINVAL;
 973                         }
 974                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
 975                 }
 976         } else if (mddev->pers == NULL) {
 977                 /* Insist of good event counter while assembling */
 978                 __u64 ev1 = le64_to_cpu(sb->events);
 979                 ++ev1;
 980                 if (ev1 < mddev->events)
 981                         return -EINVAL;
 982         } else if (mddev->bitmap) {
 983                 /* If adding to array with a bitmap, then we can accept an
 984                  * older device, but not too old.
 985                  */
 986                 __u64 ev1 = le64_to_cpu(sb->events);
 987                 if (ev1 < mddev->bitmap->events_cleared)
 988                         return 0;
 989         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 990                 return 0;
 991
 992         if (mddev->level != LEVEL_MULTIPATH) {
 993                 int role;
 994                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
 995                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
 996                 switch(role) {
 997                 case 0xffff: /* spare */
 998                         rdev->faulty = 0;
 999                         break;
1000                 case 0xfffe: /* faulty */
1001                         rdev->faulty = 1;
1002                         break;
1003                 default:
1004                         rdev->in_sync = 1;
1005                         rdev->faulty = 0;
1006                         rdev->raid_disk = role;
1007                         break;
1008                 }
1009                 rdev->flags = 0;
1010                 if (sb->devflags & WriteMostly1)
1011                         set_bit(WriteMostly, &rdev->flags);
1012         } else /* MULTIPATH are always insync */
1013                 rdev->in_sync = 1;
1014
1015         return 0;
1016 }
1017
1018 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1019 {
1020         struct mdp_superblock_1 *sb;
1021         struct list_head *tmp;
1022         mdk_rdev_t *rdev2;
1023         int max_dev, i;
1024         /* make rdev->sb match mddev and rdev data. */
1025
1026         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1027
1028         sb->feature_map = 0;
1029         sb->pad0 = 0;
1030         memset(sb->pad1, 0, sizeof(sb->pad1));
1031         memset(sb->pad2, 0, sizeof(sb->pad2));
1032         memset(sb->pad3, 0, sizeof(sb->pad3));
1033
1034         sb->utime = cpu_to_le64((__u64)mddev->utime);
1035         sb->events = cpu_to_le64(mddev->events);
1036         if (mddev->in_sync)
1037                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1038         else
1039                 sb->resync_offset = cpu_to_le64(0);
1040
1041         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1042                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1043                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1044         }
1045
1046         max_dev = 0;
1047         ITERATE_RDEV(mddev,rdev2,tmp)
1048                 if (rdev2->desc_nr+1 > max_dev)
1049                         max_dev = rdev2->desc_nr+1;
1050
1051         sb->max_dev = cpu_to_le32(max_dev);
1052         for (i=0; i<max_dev;i++)
1053                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1054
1055         ITERATE_RDEV(mddev,rdev2,tmp) {
1056                 i = rdev2->desc_nr;
1057                 if (rdev2->faulty)
1058                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1059                 else if (rdev2->in_sync)
1060                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1061                 else
1062                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1063         }
1064
1065         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1066         sb->sb_csum = calc_sb_1_csum(sb);
1067 }
1068
1069
1070 static struct super_type super_types[] = {
1071         [0] = {
1072                 .name   = "0.90.0",
1073                 .owner  = THIS_MODULE,
1074                 .load_super     = super_90_load,
1075                 .validate_super = super_90_validate,
1076                 .sync_super     = super_90_sync,
1077         },
1078         [1] = {
1079                 .name   = "md-1",
1080                 .owner  = THIS_MODULE,
1081                 .load_super     = super_1_load,
1082                 .validate_super = super_1_validate,
1083                 .sync_super     = super_1_sync,
1084         },
1085 };
1086
1087 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1088 {
1089         struct list_head *tmp;
1090         mdk_rdev_t *rdev;
1091
1092         ITERATE_RDEV(mddev,rdev,tmp)
1093                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1094                         return rdev;
1095
1096         return NULL;
1097 }
1098
1099 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1100 {
1101         struct list_head *tmp;
1102         mdk_rdev_t *rdev;
1103
1104         ITERATE_RDEV(mddev1,rdev,tmp)
1105                 if (match_dev_unit(mddev2, rdev))
1106                         return 1;
1107
1108         return 0;
1109 }
1110
1111 static LIST_HEAD(pending_raid_disks);
1112
1113 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1114 {
1115         mdk_rdev_t *same_pdev;
1116         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1117
1118         if (rdev->mddev) {
1119                 MD_BUG();
1120                 return -EINVAL;
1121         }
1122         same_pdev = match_dev_unit(mddev, rdev);
1123         if (same_pdev)
1124                 printk(KERN_WARNING
1125                         "%s: WARNING: %s appears to be on the same physical"
1126                         " disk as %s. True\n     protection against single-disk"
1127                         " failure might be compromised.\n",
1128                         mdname(mddev), bdevname(rdev->bdev,b),
1129                         bdevname(same_pdev->bdev,b2));
1130
1131         /* Verify rdev->desc_nr is unique.
1132          * If it is -1, assign a free number, else
1133          * check number is not in use
1134          */
1135         if (rdev->desc_nr < 0) {
1136                 int choice = 0;
1137                 if (mddev->pers) choice = mddev->raid_disks;
1138                 while (find_rdev_nr(mddev, choice))
1139                         choice++;
1140                 rdev->desc_nr = choice;
1141         } else {
1142                 if (find_rdev_nr(mddev, rdev->desc_nr))
1143                         return -EBUSY;
1144         }
1145
1146         list_add(&rdev->same_set, &mddev->disks);
1147         rdev->mddev = mddev;
1148         printk(KERN_INFO "md: bind<%s>\n", bdevname(rdev->bdev,b));
1149         return 0;
1150 }
1151
1152 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1153 {
1154         char b[BDEVNAME_SIZE];
1155         if (!rdev->mddev) {
1156                 MD_BUG();
1157                 return;
1158         }
1159         list_del_init(&rdev->same_set);
1160         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1161         rdev->mddev = NULL;
1162 }
1163
1164 /*
1165  * prevent the device from being mounted, repartitioned or
1166  * otherwise reused by a RAID array (or any other kernel
1167  * subsystem), by bd_claiming the device.
1168  */
1169 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1170 {
1171         int err = 0;
1172         struct block_device *bdev;
1173         char b[BDEVNAME_SIZE];
1174
1175         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1176         if (IS_ERR(bdev)) {
1177                 printk(KERN_ERR "md: could not open %s.\n",
1178                         __bdevname(dev, b));
1179                 return PTR_ERR(bdev);
1180         }
1181         err = bd_claim(bdev, rdev);
1182         if (err) {
1183                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1184                         bdevname(bdev, b));
1185                 blkdev_put(bdev);
1186                 return err;
1187         }
1188         rdev->bdev = bdev;
1189         return err;
1190 }
1191
1192 static void unlock_rdev(mdk_rdev_t *rdev)
1193 {
1194         struct block_device *bdev = rdev->bdev;
1195         rdev->bdev = NULL;
1196         if (!bdev)
1197                 MD_BUG();
1198         bd_release(bdev);
1199         blkdev_put(bdev);
1200 }
1201
1202 void md_autodetect_dev(dev_t dev);
1203
1204 static void export_rdev(mdk_rdev_t * rdev)
1205 {
1206         char b[BDEVNAME_SIZE];
1207         printk(KERN_INFO "md: export_rdev(%s)\n",
1208                 bdevname(rdev->bdev,b));
1209         if (rdev->mddev)
1210                 MD_BUG();
1211         free_disk_sb(rdev);
1212         list_del_init(&rdev->same_set);
1213 #ifndef MODULE
1214         md_autodetect_dev(rdev->bdev->bd_dev);
1215 #endif
1216         unlock_rdev(rdev);
1217         kfree(rdev);
1218 }
1219
1220 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1221 {
1222         unbind_rdev_from_array(rdev);
1223         export_rdev(rdev);
1224 }
1225
1226 static void export_array(mddev_t *mddev)
1227 {
1228         struct list_head *tmp;
1229         mdk_rdev_t *rdev;
1230
1231         ITERATE_RDEV(mddev,rdev,tmp) {
1232                 if (!rdev->mddev) {
1233                         MD_BUG();
1234                         continue;
1235                 }
1236                 kick_rdev_from_array(rdev);
1237         }
1238         if (!list_empty(&mddev->disks))
1239                 MD_BUG();
1240         mddev->raid_disks = 0;
1241         mddev->major_version = 0;
1242 }
1243
1244 static void print_desc(mdp_disk_t *desc)
1245 {
1246         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1247                 desc->major,desc->minor,desc->raid_disk,desc->state);
1248 }
1249
1250 static void print_sb(mdp_super_t *sb)
1251 {
1252         int i;
1253
1254         printk(KERN_INFO
1255                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1256                 sb->major_version, sb->minor_version, sb->patch_version,
1257                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1258                 sb->ctime);
1259         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1260                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1261                 sb->md_minor, sb->layout, sb->chunk_size);
1262         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1263                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1264                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1265                 sb->failed_disks, sb->spare_disks,
1266                 sb->sb_csum, (unsigned long)sb->events_lo);
1267
1268         printk(KERN_INFO);
1269         for (i = 0; i < MD_SB_DISKS; i++) {
1270                 mdp_disk_t *desc;
1271
1272                 desc = sb->disks + i;
1273                 if (desc->number || desc->major || desc->minor ||
1274                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1275                         printk("     D %2d: ", i);
1276                         print_desc(desc);
1277                 }
1278         }
1279         printk(KERN_INFO "md:     THIS: ");
1280         print_desc(&sb->this_disk);
1281
1282 }
1283
1284 static void print_rdev(mdk_rdev_t *rdev)
1285 {
1286         char b[BDEVNAME_SIZE];
1287         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1288                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1289                 rdev->faulty, rdev->in_sync, rdev->desc_nr);
1290         if (rdev->sb_loaded) {
1291                 printk(KERN_INFO "md: rdev superblock:\n");
1292                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1293         } else
1294                 printk(KERN_INFO "md: no rdev superblock!\n");
1295 }
1296
1297 void md_print_devices(void)
1298 {
1299         struct list_head *tmp, *tmp2;
1300         mdk_rdev_t *rdev;
1301         mddev_t *mddev;
1302         char b[BDEVNAME_SIZE];
1303
1304         printk("\n");
1305         printk("md:     **********************************\n");
1306         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1307         printk("md:     **********************************\n");
1308         ITERATE_MDDEV(mddev,tmp) {
1309
1310                 if (mddev->bitmap)
1311                         bitmap_print_sb(mddev->bitmap);
1312                 else
1313                         printk("%s: ", mdname(mddev));
1314                 ITERATE_RDEV(mddev,rdev,tmp2)
1315                         printk("<%s>", bdevname(rdev->bdev,b));
1316                 printk("\n");
1317
1318                 ITERATE_RDEV(mddev,rdev,tmp2)
1319                         print_rdev(rdev);
1320         }
1321         printk("md:     **********************************\n");
1322         printk("\n");
1323 }
1324
1325
1326 static void sync_sbs(mddev_t * mddev)
1327 {
1328         mdk_rdev_t *rdev;
1329         struct list_head *tmp;
1330
1331         ITERATE_RDEV(mddev,rdev,tmp) {
1332                 super_types[mddev->major_version].
1333                         sync_super(mddev, rdev);
1334                 rdev->sb_loaded = 1;
1335         }
1336 }
1337
1338 static void md_update_sb(mddev_t * mddev)
1339 {
1340         int err;
1341         struct list_head *tmp;
1342         mdk_rdev_t *rdev;
1343         int sync_req;
1344
1345 repeat:
1346         spin_lock(&mddev->write_lock);
1347         sync_req = mddev->in_sync;
1348         mddev->utime = get_seconds();
1349         mddev->events ++;
1350
1351         if (!mddev->events) {
1352                 /*
1353                  * oops, this 64-bit counter should never wrap.
1354                  * Either we are in around ~1 trillion A.C., assuming
1355                  * 1 reboot per second, or we have a bug:
1356                  */
1357                 MD_BUG();
1358                 mddev->events --;
1359         }
1360         mddev->sb_dirty = 2;
1361         sync_sbs(mddev);
1362
1363         /*
1364          * do not write anything to disk if using
1365          * nonpersistent superblocks
1366          */
1367         if (!mddev->persistent) {
1368                 mddev->sb_dirty = 0;
1369                 spin_unlock(&mddev->write_lock);
1370                 wake_up(&mddev->sb_wait);
1371                 return;
1372         }
1373         spin_unlock(&mddev->write_lock);
1374
1375         dprintk(KERN_INFO
1376                 "md: updating %s RAID superblock on device (in sync %d)\n",
1377                 mdname(mddev),mddev->in_sync);
1378
1379         err = bitmap_update_sb(mddev->bitmap);
1380         ITERATE_RDEV(mddev,rdev,tmp) {
1381                 char b[BDEVNAME_SIZE];
1382                 dprintk(KERN_INFO "md: ");
1383                 if (rdev->faulty)
1384                         dprintk("(skipping faulty ");
1385
1386                 dprintk("%s ", bdevname(rdev->bdev,b));
1387                 if (!rdev->faulty) {
1388                         md_super_write(mddev,rdev,
1389                                        rdev->sb_offset<<1, rdev->sb_size,
1390                                        rdev->sb_page);
1391                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1392                                 bdevname(rdev->bdev,b),
1393                                 (unsigned long long)rdev->sb_offset);
1394
1395                 } else
1396                         dprintk(")\n");
1397                 if (mddev->level == LEVEL_MULTIPATH)
1398                         /* only need to write one superblock... */
1399                         break;
1400         }
1401         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1402         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1403
1404         spin_lock(&mddev->write_lock);
1405         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1406                 /* have to write it out again */
1407                 spin_unlock(&mddev->write_lock);
1408                 goto repeat;
1409         }
1410         mddev->sb_dirty = 0;
1411         spin_unlock(&mddev->write_lock);
1412         wake_up(&mddev->sb_wait);
1413
1414 }
1415
1416 /*
1417  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1418  *
1419  * mark the device faulty if:
1420  *
1421  *   - the device is nonexistent (zero size)
1422  *   - the device has no valid superblock
1423  *
1424  * a faulty rdev _never_ has rdev->sb set.
1425  */
1426 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1427 {
1428         char b[BDEVNAME_SIZE];
1429         int err;
1430         mdk_rdev_t *rdev;
1431         sector_t size;
1432
1433         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1434         if (!rdev) {
1435                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1436                 return ERR_PTR(-ENOMEM);
1437         }
1438         memset(rdev, 0, sizeof(*rdev));
1439
1440         if ((err = alloc_disk_sb(rdev)))
1441                 goto abort_free;
1442
1443         err = lock_rdev(rdev, newdev);
1444         if (err)
1445                 goto abort_free;
1446
1447         rdev->desc_nr = -1;
1448         rdev->faulty = 0;
1449         rdev->in_sync = 0;
1450         rdev->data_offset = 0;
1451         atomic_set(&rdev->nr_pending, 0);
1452
1453         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1454         if (!size) {
1455                 printk(KERN_WARNING
1456                         "md: %s has zero or unknown size, marking faulty!\n",
1457                         bdevname(rdev->bdev,b));
1458                 err = -EINVAL;
1459                 goto abort_free;
1460         }
1461
1462         if (super_format >= 0) {
1463                 err = super_types[super_format].
1464                         load_super(rdev, NULL, super_minor);
1465                 if (err == -EINVAL) {
1466                         printk(KERN_WARNING
1467                                 "md: %s has invalid sb, not importing!\n",
1468                                 bdevname(rdev->bdev,b));
1469                         goto abort_free;
1470                 }
1471                 if (err < 0) {
1472                         printk(KERN_WARNING
1473                                 "md: could not read %s's sb, not importing!\n",
1474                                 bdevname(rdev->bdev,b));
1475                         goto abort_free;
1476                 }
1477         }
1478         INIT_LIST_HEAD(&rdev->same_set);
1479
1480         return rdev;
1481
1482 abort_free:
1483         if (rdev->sb_page) {
1484                 if (rdev->bdev)
1485                         unlock_rdev(rdev);
1486                 free_disk_sb(rdev);
1487         }
1488         kfree(rdev);
1489         return ERR_PTR(err);
1490 }
1491
1492 /*
1493  * Check a full RAID array for plausibility
1494  */
1495
1496
1497 static void analyze_sbs(mddev_t * mddev)
1498 {
1499         int i;
1500         struct list_head *tmp;
1501         mdk_rdev_t *rdev, *freshest;
1502         char b[BDEVNAME_SIZE];
1503
1504         freshest = NULL;
1505         ITERATE_RDEV(mddev,rdev,tmp)
1506                 switch (super_types[mddev->major_version].
1507                         load_super(rdev, freshest, mddev->minor_version)) {
1508                 case 1:
1509                         freshest = rdev;
1510                         break;
1511                 case 0:
1512                         break;
1513                 default:
1514                         printk( KERN_ERR \
1515                                 "md: fatal superblock inconsistency in %s"
1516                                 " -- removing from array\n",
1517                                 bdevname(rdev->bdev,b));
1518                         kick_rdev_from_array(rdev);
1519                 }
1520
1521
1522         super_types[mddev->major_version].
1523                 validate_super(mddev, freshest);
1524
1525         i = 0;
1526         ITERATE_RDEV(mddev,rdev,tmp) {
1527                 if (rdev != freshest)
1528                         if (super_types[mddev->major_version].
1529                             validate_super(mddev, rdev)) {
1530                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1531                                         " from array!\n",
1532                                         bdevname(rdev->bdev,b));
1533                                 kick_rdev_from_array(rdev);
1534                                 continue;
1535                         }
1536                 if (mddev->level == LEVEL_MULTIPATH) {
1537                         rdev->desc_nr = i++;
1538                         rdev->raid_disk = rdev->desc_nr;
1539                         rdev->in_sync = 1;
1540                 }
1541         }
1542
1543
1544
1545         if (mddev->recovery_cp != MaxSector &&
1546             mddev->level >= 1)
1547                 printk(KERN_ERR "md: %s: raid array is not clean"
1548                        " -- starting background reconstruction\n",
1549                        mdname(mddev));
1550
1551 }
1552
1553 int mdp_major = 0;
1554
1555 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1556 {
1557         static DECLARE_MUTEX(disks_sem);
1558         mddev_t *mddev = mddev_find(dev);
1559         struct gendisk *disk;
1560         int partitioned = (MAJOR(dev) != MD_MAJOR);
1561         int shift = partitioned ? MdpMinorShift : 0;
1562         int unit = MINOR(dev) >> shift;
1563
1564         if (!mddev)
1565                 return NULL;
1566
1567         down(&disks_sem);
1568         if (mddev->gendisk) {
1569                 up(&disks_sem);
1570                 mddev_put(mddev);
1571                 return NULL;
1572         }
1573         disk = alloc_disk(1 << shift);
1574         if (!disk) {
1575                 up(&disks_sem);
1576                 mddev_put(mddev);
1577                 return NULL;
1578         }
1579         disk->major = MAJOR(dev);
1580         disk->first_minor = unit << shift;
1581         if (partitioned) {
1582                 sprintf(disk->disk_name, "md_d%d", unit);
1583                 sprintf(disk->devfs_name, "md/d%d", unit);
1584         } else {
1585                 sprintf(disk->disk_name, "md%d", unit);
1586                 sprintf(disk->devfs_name, "md/%d", unit);
1587         }
1588         disk->fops = &md_fops;
1589         disk->private_data = mddev;
1590         disk->queue = mddev->queue;
1591         add_disk(disk);
1592         mddev->gendisk = disk;
1593         up(&disks_sem);
1594         return NULL;
1595 }
1596
1597 void md_wakeup_thread(mdk_thread_t *thread);
1598
1599 static void md_safemode_timeout(unsigned long data)
1600 {
1601         mddev_t *mddev = (mddev_t *) data;
1602
1603         mddev->safemode = 1;
1604         md_wakeup_thread(mddev->thread);
1605 }
1606
1607
1608 static int do_md_run(mddev_t * mddev)
1609 {
1610         int pnum, err;
1611         int chunk_size;
1612         struct list_head *tmp;
1613         mdk_rdev_t *rdev;
1614         struct gendisk *disk;
1615         char b[BDEVNAME_SIZE];
1616
1617         if (list_empty(&mddev->disks))
1618                 /* cannot run an array with no devices.. */
1619                 return -EINVAL;
1620
1621         if (mddev->pers)
1622                 return -EBUSY;
1623
1624         /*
1625          * Analyze all RAID superblock(s)
1626          */
1627         if (!mddev->raid_disks)
1628                 analyze_sbs(mddev);
1629
1630         chunk_size = mddev->chunk_size;
1631         pnum = level_to_pers(mddev->level);
1632
1633         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1634                 if (!chunk_size) {
1635                         /*
1636                          * 'default chunksize' in the old md code used to
1637                          * be PAGE_SIZE, baaad.
1638                          * we abort here to be on the safe side. We don't
1639                          * want to continue the bad practice.
1640                          */
1641                         printk(KERN_ERR
1642                                 "no chunksize specified, see 'man raidtab'\n");
1643                         return -EINVAL;
1644                 }
1645                 if (chunk_size > MAX_CHUNK_SIZE) {
1646                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1647                                 chunk_size, MAX_CHUNK_SIZE);
1648                         return -EINVAL;
1649                 }
1650                 /*
1651                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1652                  */
1653                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1654                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1655                         return -EINVAL;
1656                 }
1657                 if (chunk_size < PAGE_SIZE) {
1658                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1659                                 chunk_size, PAGE_SIZE);
1660                         return -EINVAL;
1661                 }
1662
1663                 /* devices must have minimum size of one chunk */
1664                 ITERATE_RDEV(mddev,rdev,tmp) {
1665                         if (rdev->faulty)
1666                                 continue;
1667                         if (rdev->size < chunk_size / 1024) {
1668                                 printk(KERN_WARNING
1669                                         "md: Dev %s smaller than chunk_size:"
1670                                         " %lluk < %dk\n",
1671                                         bdevname(rdev->bdev,b),
1672                                         (unsigned long long)rdev->size,
1673                                         chunk_size / 1024);
1674                                 return -EINVAL;
1675                         }
1676                 }
1677         }
1678
1679 #ifdef CONFIG_KMOD
1680         if (!pers[pnum])
1681         {
1682                 request_module("md-personality-%d", pnum);
1683         }
1684 #endif
1685
1686         /*
1687          * Drop all container device buffers, from now on
1688          * the only valid external interface is through the md
1689          * device.
1690          * Also find largest hardsector size
1691          */
1692         ITERATE_RDEV(mddev,rdev,tmp) {
1693                 if (rdev->faulty)
1694                         continue;
1695                 sync_blockdev(rdev->bdev);
1696                 invalidate_bdev(rdev->bdev, 0);
1697         }
1698
1699         md_probe(mddev->unit, NULL, NULL);
1700         disk = mddev->gendisk;
1701         if (!disk)
1702                 return -ENOMEM;
1703
1704         spin_lock(&pers_lock);
1705         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
1706                 spin_unlock(&pers_lock);
1707                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
1708                        pnum);
1709                 return -EINVAL;
1710         }
1711
1712         mddev->pers = pers[pnum];
1713         spin_unlock(&pers_lock);
1714
1715         mddev->recovery = 0;
1716         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
1717
1718         /* before we start the array running, initialise the bitmap */
1719         err = bitmap_create(mddev);
1720         if (err)
1721                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
1722                         mdname(mddev), err);
1723         else
1724                 err = mddev->pers->run(mddev);
1725         if (err) {
1726                 printk(KERN_ERR "md: pers->run() failed ...\n");
1727                 module_put(mddev->pers->owner);
1728                 mddev->pers = NULL;
1729                 bitmap_destroy(mddev);
1730                 return err;
1731         }
1732         atomic_set(&mddev->writes_pending,0);
1733         mddev->safemode = 0;
1734         mddev->safemode_timer.function = md_safemode_timeout;
1735         mddev->safemode_timer.data = (unsigned long) mddev;
1736         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
1737         mddev->in_sync = 1;
1738
1739         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1740         md_wakeup_thread(mddev->thread);
1741
1742         if (mddev->sb_dirty)
1743                 md_update_sb(mddev);
1744
1745         set_capacity(disk, mddev->array_size<<1);
1746
1747         /* If we call blk_queue_make_request here, it will
1748          * re-initialise max_sectors etc which may have been
1749          * refined inside -> run.  So just set the bits we need to set.
1750          * Most initialisation happended when we called
1751          * blk_queue_make_request(..., md_fail_request)
1752          * earlier.
1753          */
1754         mddev->queue->queuedata = mddev;
1755         mddev->queue->make_request_fn = mddev->pers->make_request;
1756
1757         mddev->changed = 1;
1758         return 0;
1759 }
1760
1761 static int restart_array(mddev_t *mddev)
1762 {
1763         struct gendisk *disk = mddev->gendisk;
1764         int err;
1765
1766         /*
1767          * Complain if it has no devices
1768          */
1769         err = -ENXIO;
1770         if (list_empty(&mddev->disks))
1771                 goto out;
1772
1773         if (mddev->pers) {
1774                 err = -EBUSY;
1775                 if (!mddev->ro)
1776                         goto out;
1777
1778                 mddev->safemode = 0;
1779                 mddev->ro = 0;
1780                 set_disk_ro(disk, 0);
1781
1782                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
1783                         mdname(mddev));
1784                 /*
1785                  * Kick recovery or resync if necessary
1786                  */
1787                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1788                 md_wakeup_thread(mddev->thread);
1789                 err = 0;
1790         } else {
1791                 printk(KERN_ERR "md: %s has no personality assigned.\n",
1792                         mdname(mddev));
1793                 err = -EINVAL;
1794         }
1795
1796 out:
1797         return err;
1798 }
1799
1800 static int do_md_stop(mddev_t * mddev, int ro)
1801 {
1802         int err = 0;
1803         struct gendisk *disk = mddev->gendisk;
1804
1805         if (mddev->pers) {
1806                 if (atomic_read(&mddev->active)>2) {
1807                         printk("md: %s still in use.\n",mdname(mddev));
1808                         return -EBUSY;
1809                 }
1810
1811                 if (mddev->sync_thread) {
1812                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1813                         md_unregister_thread(mddev->sync_thread);
1814                         mddev->sync_thread = NULL;
1815                 }
1816
1817                 del_timer_sync(&mddev->safemode_timer);
1818
1819                 invalidate_partition(disk, 0);
1820
1821                 if (ro) {
1822                         err  = -ENXIO;
1823                         if (mddev->ro)
1824                                 goto out;
1825                         mddev->ro = 1;
1826                 } else {
1827                         bitmap_flush(mddev);
1828                         wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
1829                         if (mddev->ro)
1830                                 set_disk_ro(disk, 0);
1831                         blk_queue_make_request(mddev->queue, md_fail_request);
1832                         mddev->pers->stop(mddev);
1833                         module_put(mddev->pers->owner);
1834                         mddev->pers = NULL;
1835                         if (mddev->ro)
1836                                 mddev->ro = 0;
1837                 }
1838                 if (!mddev->in_sync) {
1839                         /* mark array as shutdown cleanly */
1840                         mddev->in_sync = 1;
1841                         md_update_sb(mddev);
1842                 }
1843                 if (ro)
1844                         set_disk_ro(disk, 1);
1845         }
1846
1847         bitmap_destroy(mddev);
1848         if (mddev->bitmap_file) {
1849                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
1850                 fput(mddev->bitmap_file);
1851                 mddev->bitmap_file = NULL;
1852         }
1853         mddev->bitmap_offset = 0;
1854
1855         /*
1856          * Free resources if final stop
1857          */
1858         if (!ro) {
1859                 struct gendisk *disk;
1860                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
1861
1862                 export_array(mddev);
1863
1864                 mddev->array_size = 0;
1865                 disk = mddev->gendisk;
1866                 if (disk)
1867                         set_capacity(disk, 0);
1868                 mddev->changed = 1;
1869         } else
1870                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
1871                         mdname(mddev));
1872         err = 0;
1873 out:
1874         return err;
1875 }
1876
1877 static void autorun_array(mddev_t *mddev)
1878 {
1879         mdk_rdev_t *rdev;
1880         struct list_head *tmp;
1881         int err;
1882
1883         if (list_empty(&mddev->disks))
1884                 return;
1885
1886         printk(KERN_INFO "md: running: ");
1887
1888         ITERATE_RDEV(mddev,rdev,tmp) {
1889                 char b[BDEVNAME_SIZE];
1890                 printk("<%s>", bdevname(rdev->bdev,b));
1891         }
1892         printk("\n");
1893
1894         err = do_md_run (mddev);
1895         if (err) {
1896                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
1897                 do_md_stop (mddev, 0);
1898         }
1899 }
1900
1901 /*
1902  * lets try to run arrays based on all disks that have arrived
1903  * until now. (those are in pending_raid_disks)
1904  *
1905  * the method: pick the first pending disk, collect all disks with
1906  * the same UUID, remove all from the pending list and put them into
1907  * the 'same_array' list. Then order this list based on superblock
1908  * update time (freshest comes first), kick out 'old' disks and
1909  * compare superblocks. If everything's fine then run it.
1910  *
1911  * If "unit" is allocated, then bump its reference count
1912  */
1913 static void autorun_devices(int part)
1914 {
1915         struct list_head candidates;
1916         struct list_head *tmp;
1917         mdk_rdev_t *rdev0, *rdev;
1918         mddev_t *mddev;
1919         char b[BDEVNAME_SIZE];
1920
1921         printk(KERN_INFO "md: autorun ...\n");
1922         while (!list_empty(&pending_raid_disks)) {
1923                 dev_t dev;
1924                 rdev0 = list_entry(pending_raid_disks.next,
1925                                          mdk_rdev_t, same_set);
1926
1927                 printk(KERN_INFO "md: considering %s ...\n",
1928                         bdevname(rdev0->bdev,b));
1929                 INIT_LIST_HEAD(&candidates);
1930                 ITERATE_RDEV_PENDING(rdev,tmp)
1931                         if (super_90_load(rdev, rdev0, 0) >= 0) {
1932                                 printk(KERN_INFO "md:  adding %s ...\n",
1933                                         bdevname(rdev->bdev,b));
1934                                 list_move(&rdev->same_set, &candidates);
1935                         }
1936                 /*
1937                  * now we have a set of devices, with all of them having
1938                  * mostly sane superblocks. It's time to allocate the
1939                  * mddev.
1940                  */
1941                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
1942                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
1943                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
1944                         break;
1945                 }
1946                 if (part)
1947                         dev = MKDEV(mdp_major,
1948                                     rdev0->preferred_minor << MdpMinorShift);
1949                 else
1950                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
1951
1952                 md_probe(dev, NULL, NULL);
1953                 mddev = mddev_find(dev);
1954                 if (!mddev) {
1955                         printk(KERN_ERR
1956                                 "md: cannot allocate memory for md drive.\n");
1957                         break;
1958                 }
1959                 if (mddev_lock(mddev))
1960                         printk(KERN_WARNING "md: %s locked, cannot run\n",
1961                                mdname(mddev));
1962                 else if (mddev->raid_disks || mddev->major_version
1963                          || !list_empty(&mddev->disks)) {
1964                         printk(KERN_WARNING
1965                                 "md: %s already running, cannot run %s\n",
1966                                 mdname(mddev), bdevname(rdev0->bdev,b));
1967                         mddev_unlock(mddev);
1968                 } else {
1969                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
1970                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
1971                                 list_del_init(&rdev->same_set);
1972                                 if (bind_rdev_to_array(rdev, mddev))
1973                                         export_rdev(rdev);
1974                         }
1975                         autorun_array(mddev);
1976                         mddev_unlock(mddev);
1977                 }
1978                 /* on success, candidates will be empty, on error
1979                  * it won't...
1980                  */
1981                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
1982                         export_rdev(rdev);
1983                 mddev_put(mddev);
1984         }
1985         printk(KERN_INFO "md: ... autorun DONE.\n");
1986 }
1987
1988 /*
1989  * import RAID devices based on one partition
1990  * if possible, the array gets run as well.
1991  */
1992
1993 static int autostart_array(dev_t startdev)
1994 {
1995         char b[BDEVNAME_SIZE];
1996         int err = -EINVAL, i;
1997         mdp_super_t *sb = NULL;
1998         mdk_rdev_t *start_rdev = NULL, *rdev;
1999
2000         start_rdev = md_import_device(startdev, 0, 0);
2001         if (IS_ERR(start_rdev))
2002                 return err;
2003
2004
2005         /* NOTE: this can only work for 0.90.0 superblocks */
2006         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2007         if (sb->major_version != 0 ||
2008             sb->minor_version != 90 ) {
2009                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2010                 export_rdev(start_rdev);
2011                 return err;
2012         }
2013
2014         if (start_rdev->faulty) {
2015                 printk(KERN_WARNING
2016                         "md: can not autostart based on faulty %s!\n",
2017                         bdevname(start_rdev->bdev,b));
2018                 export_rdev(start_rdev);
2019                 return err;
2020         }
2021         list_add(&start_rdev->same_set, &pending_raid_disks);
2022
2023         for (i = 0; i < MD_SB_DISKS; i++) {
2024                 mdp_disk_t *desc = sb->disks + i;
2025                 dev_t dev = MKDEV(desc->major, desc->minor);
2026
2027                 if (!dev)
2028                         continue;
2029                 if (dev == startdev)
2030                         continue;
2031                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2032                         continue;
2033                 rdev = md_import_device(dev, 0, 0);
2034                 if (IS_ERR(rdev))
2035                         continue;
2036
2037                 list_add(&rdev->same_set, &pending_raid_disks);
2038         }
2039
2040         /*
2041          * possibly return codes
2042          */
2043         autorun_devices(0);
2044         return 0;
2045
2046 }
2047
2048
2049 static int get_version(void __user * arg)
2050 {
2051         mdu_version_t ver;
2052
2053         ver.major = MD_MAJOR_VERSION;
2054         ver.minor = MD_MINOR_VERSION;
2055         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2056
2057         if (copy_to_user(arg, &ver, sizeof(ver)))
2058                 return -EFAULT;
2059
2060         return 0;
2061 }
2062
2063 static int get_array_info(mddev_t * mddev, void __user * arg)
2064 {
2065         mdu_array_info_t info;
2066         int nr,working,active,failed,spare;
2067         mdk_rdev_t *rdev;
2068         struct list_head *tmp;
2069
2070         nr=working=active=failed=spare=0;
2071         ITERATE_RDEV(mddev,rdev,tmp) {
2072                 nr++;
2073                 if (rdev->faulty)
2074                         failed++;
2075                 else {
2076                         working++;
2077                         if (rdev->in_sync)
2078                                 active++;
2079                         else
2080                                 spare++;
2081                 }
2082         }
2083
2084         info.major_version = mddev->major_version;
2085         info.minor_version = mddev->minor_version;
2086         info.patch_version = MD_PATCHLEVEL_VERSION;
2087         info.ctime         = mddev->ctime;
2088         info.level         = mddev->level;
2089         info.size          = mddev->size;
2090         info.nr_disks      = nr;
2091         info.raid_disks    = mddev->raid_disks;
2092         info.md_minor      = mddev->md_minor;
2093         info.not_persistent= !mddev->persistent;
2094
2095         info.utime         = mddev->utime;
2096         info.state         = 0;
2097         if (mddev->in_sync)
2098                 info.state = (1<<MD_SB_CLEAN);
2099         if (mddev->bitmap && mddev->bitmap_offset)
2100                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2101         info.active_disks  = active;
2102         info.working_disks = working;
2103         info.failed_disks  = failed;
2104         info.spare_disks   = spare;
2105
2106         info.layout        = mddev->layout;
2107         info.chunk_size    = mddev->chunk_size;
2108
2109         if (copy_to_user(arg, &info, sizeof(info)))
2110                 return -EFAULT;
2111
2112         return 0;
2113 }
2114
2115 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2116 {
2117         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2118         char *ptr, *buf = NULL;
2119         int err = -ENOMEM;
2120
2121         file = kmalloc(sizeof(*file), GFP_KERNEL);
2122         if (!file)
2123                 goto out;
2124
2125         /* bitmap disabled, zero the first byte and copy out */
2126         if (!mddev->bitmap || !mddev->bitmap->file) {
2127                 file->pathname[0] = '\0';
2128                 goto copy_out;
2129         }
2130
2131         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2132         if (!buf)
2133                 goto out;
2134
2135         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2136         if (!ptr)
2137                 goto out;
2138
2139         strcpy(file->pathname, ptr);
2140
2141 copy_out:
2142         err = 0;
2143         if (copy_to_user(arg, file, sizeof(*file)))
2144                 err = -EFAULT;
2145 out:
2146         kfree(buf);
2147         kfree(file);
2148         return err;
2149 }
2150
2151 static int get_disk_info(mddev_t * mddev, void __user * arg)
2152 {
2153         mdu_disk_info_t info;
2154         unsigned int nr;
2155         mdk_rdev_t *rdev;
2156
2157         if (copy_from_user(&info, arg, sizeof(info)))
2158                 return -EFAULT;
2159
2160         nr = info.number;
2161
2162         rdev = find_rdev_nr(mddev, nr);
2163         if (rdev) {
2164                 info.major = MAJOR(rdev->bdev->bd_dev);
2165                 info.minor = MINOR(rdev->bdev->bd_dev);
2166                 info.raid_disk = rdev->raid_disk;
2167                 info.state = 0;
2168                 if (rdev->faulty)
2169                         info.state |= (1<<MD_DISK_FAULTY);
2170                 else if (rdev->in_sync) {
2171                         info.state |= (1<<MD_DISK_ACTIVE);
2172                         info.state |= (1<<MD_DISK_SYNC);
2173                 }
2174                 if (test_bit(WriteMostly, &rdev->flags))
2175                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2176         } else {
2177                 info.major = info.minor = 0;
2178                 info.raid_disk = -1;
2179                 info.state = (1<<MD_DISK_REMOVED);
2180         }
2181
2182         if (copy_to_user(arg, &info, sizeof(info)))
2183                 return -EFAULT;
2184
2185         return 0;
2186 }
2187
2188 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2189 {
2190         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2191         mdk_rdev_t *rdev;
2192         dev_t dev = MKDEV(info->major,info->minor);
2193
2194         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2195                 return -EOVERFLOW;
2196
2197         if (!mddev->raid_disks) {
2198                 int err;
2199                 /* expecting a device which has a superblock */
2200                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2201                 if (IS_ERR(rdev)) {
2202                         printk(KERN_WARNING
2203                                 "md: md_import_device returned %ld\n",
2204                                 PTR_ERR(rdev));
2205                         return PTR_ERR(rdev);
2206                 }
2207                 if (!list_empty(&mddev->disks)) {
2208                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2209                                                         mdk_rdev_t, same_set);
2210                         int err = super_types[mddev->major_version]
2211                                 .load_super(rdev, rdev0, mddev->minor_version);
2212                         if (err < 0) {
2213                                 printk(KERN_WARNING
2214                                         "md: %s has different UUID to %s\n",
2215                                         bdevname(rdev->bdev,b),
2216                                         bdevname(rdev0->bdev,b2));
2217                                 export_rdev(rdev);
2218                                 return -EINVAL;
2219                         }
2220                 }
2221                 err = bind_rdev_to_array(rdev, mddev);
2222                 if (err)
2223                         export_rdev(rdev);
2224                 return err;
2225         }
2226
2227         /*
2228          * add_new_disk can be used once the array is assembled
2229          * to add "hot spares".  They must already have a superblock
2230          * written
2231          */
2232         if (mddev->pers) {
2233                 int err;
2234                 if (!mddev->pers->hot_add_disk) {
2235                         printk(KERN_WARNING
2236                                 "%s: personality does not support diskops!\n",
2237                                mdname(mddev));
2238                         return -EINVAL;
2239                 }
2240                 if (mddev->persistent)
2241                         rdev = md_import_device(dev, mddev->major_version,
2242                                                 mddev->minor_version);
2243                 else
2244                         rdev = md_import_device(dev, -1, -1);
2245                 if (IS_ERR(rdev)) {
2246                         printk(KERN_WARNING
2247                                 "md: md_import_device returned %ld\n",
2248                                 PTR_ERR(rdev));
2249                         return PTR_ERR(rdev);
2250                 }
2251                 /* set save_raid_disk if appropriate */
2252                 if (!mddev->persistent) {
2253                         if (info->state & (1<<MD_DISK_SYNC)  &&
2254                             info->raid_disk < mddev->raid_disks)
2255                                 rdev->raid_disk = info->raid_disk;
2256                         else
2257                                 rdev->raid_disk = -1;
2258                 } else
2259                         super_types[mddev->major_version].
2260                                 validate_super(mddev, rdev);
2261                 rdev->saved_raid_disk = rdev->raid_disk;
2262
2263                 rdev->in_sync = 0; /* just to be sure */
2264                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2265                         set_bit(WriteMostly, &rdev->flags);
2266
2267                 rdev->raid_disk = -1;
2268                 err = bind_rdev_to_array(rdev, mddev);
2269                 if (err)
2270                         export_rdev(rdev);
2271
2272                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2273                 md_wakeup_thread(mddev->thread);
2274                 return err;
2275         }
2276
2277         /* otherwise, add_new_disk is only allowed
2278          * for major_version==0 superblocks
2279          */
2280         if (mddev->major_version != 0) {
2281                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2282                        mdname(mddev));
2283                 return -EINVAL;
2284         }
2285
2286         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2287                 int err;
2288                 rdev = md_import_device (dev, -1, 0);
2289                 if (IS_ERR(rdev)) {
2290                         printk(KERN_WARNING
2291                                 "md: error, md_import_device() returned %ld\n",
2292                                 PTR_ERR(rdev));
2293                         return PTR_ERR(rdev);
2294                 }
2295                 rdev->desc_nr = info->number;
2296                 if (info->raid_disk < mddev->raid_disks)
2297                         rdev->raid_disk = info->raid_disk;
2298                 else
2299                         rdev->raid_disk = -1;
2300
2301                 rdev->faulty = 0;
2302                 if (rdev->raid_disk < mddev->raid_disks)
2303                         rdev->in_sync = (info->state & (1<<MD_DISK_SYNC));
2304                 else
2305                         rdev->in_sync = 0;
2306
2307                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2308                         set_bit(WriteMostly, &rdev->flags);
2309
2310                 err = bind_rdev_to_array(rdev, mddev);
2311                 if (err) {
2312                         export_rdev(rdev);
2313                         return err;
2314                 }
2315
2316                 if (!mddev->persistent) {
2317                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2318                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2319                 } else
2320                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2321                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2322
2323                 if (!mddev->size || (mddev->size > rdev->size))
2324                         mddev->size = rdev->size;
2325         }
2326
2327         return 0;
2328 }
2329
2330 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2331 {
2332         char b[BDEVNAME_SIZE];
2333         mdk_rdev_t *rdev;
2334
2335         if (!mddev->pers)
2336                 return -ENODEV;
2337
2338         rdev = find_rdev(mddev, dev);
2339         if (!rdev)
2340                 return -ENXIO;
2341
2342         if (rdev->raid_disk >= 0)
2343                 goto busy;
2344
2345         kick_rdev_from_array(rdev);
2346         md_update_sb(mddev);
2347
2348         return 0;
2349 busy:
2350         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2351                 bdevname(rdev->bdev,b), mdname(mddev));
2352         return -EBUSY;
2353 }
2354
2355 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2356 {
2357         char b[BDEVNAME_SIZE];
2358         int err;
2359         unsigned int size;
2360         mdk_rdev_t *rdev;
2361
2362         if (!mddev->pers)
2363                 return -ENODEV;
2364
2365         if (mddev->major_version != 0) {
2366                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2367                         " version-0 superblocks.\n",
2368                         mdname(mddev));
2369                 return -EINVAL;
2370         }
2371         if (!mddev->pers->hot_add_disk) {
2372                 printk(KERN_WARNING
2373                         "%s: personality does not support diskops!\n",
2374                         mdname(mddev));
2375                 return -EINVAL;
2376         }
2377
2378         rdev = md_import_device (dev, -1, 0);
2379         if (IS_ERR(rdev)) {
2380                 printk(KERN_WARNING
2381                         "md: error, md_import_device() returned %ld\n",
2382                         PTR_ERR(rdev));
2383                 return -EINVAL;
2384         }
2385
2386         if (mddev->persistent)
2387                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2388         else
2389                 rdev->sb_offset =
2390                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2391
2392         size = calc_dev_size(rdev, mddev->chunk_size);
2393         rdev->size = size;
2394
2395         if (size < mddev->size) {
2396                 printk(KERN_WARNING
2397                         "%s: disk size %llu blocks < array size %llu\n",
2398                         mdname(mddev), (unsigned long long)size,
2399                         (unsigned long long)mddev->size);
2400                 err = -ENOSPC;
2401                 goto abort_export;
2402         }
2403
2404         if (rdev->faulty) {
2405                 printk(KERN_WARNING
2406                         "md: can not hot-add faulty %s disk to %s!\n",
2407                         bdevname(rdev->bdev,b), mdname(mddev));
2408                 err = -EINVAL;
2409                 goto abort_export;
2410         }
2411         rdev->in_sync = 0;
2412         rdev->desc_nr = -1;
2413         bind_rdev_to_array(rdev, mddev);
2414
2415         /*
2416          * The rest should better be atomic, we can have disk failures
2417          * noticed in interrupt contexts ...
2418          */
2419
2420         if (rdev->desc_nr == mddev->max_disks) {
2421                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2422                         mdname(mddev));
2423                 err = -EBUSY;
2424                 goto abort_unbind_export;
2425         }
2426
2427         rdev->raid_disk = -1;
2428
2429         md_update_sb(mddev);
2430
2431         /*
2432          * Kick recovery, maybe this spare has to be added to the
2433          * array immediately.
2434          */
2435         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2436         md_wakeup_thread(mddev->thread);
2437
2438         return 0;
2439
2440 abort_unbind_export:
2441         unbind_rdev_from_array(rdev);
2442
2443 abort_export:
2444         export_rdev(rdev);
2445         return err;
2446 }
2447
2448 /* similar to deny_write_access, but accounts for our holding a reference
2449  * to the file ourselves */
2450 static int deny_bitmap_write_access(struct file * file)
2451 {
2452         struct inode *inode = file->f_mapping->host;
2453
2454         spin_lock(&inode->i_lock);
2455         if (atomic_read(&inode->i_writecount) > 1) {
2456                 spin_unlock(&inode->i_lock);
2457                 return -ETXTBSY;
2458         }
2459         atomic_set(&inode->i_writecount, -1);
2460         spin_unlock(&inode->i_lock);
2461
2462         return 0;
2463 }
2464
2465 static int set_bitmap_file(mddev_t *mddev, int fd)
2466 {
2467         int err;
2468
2469         if (mddev->pers) {
2470                 if (!mddev->pers->quiesce)
2471                         return -EBUSY;
2472                 if (mddev->recovery || mddev->sync_thread)
2473                         return -EBUSY;
2474                 /* we should be able to change the bitmap.. */
2475         }
2476
2477
2478         if (fd >= 0) {
2479                 if (mddev->bitmap)
2480                         return -EEXIST; /* cannot add when bitmap is present */
2481                 mddev->bitmap_file = fget(fd);
2482
2483                 if (mddev->bitmap_file == NULL) {
2484                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2485                                mdname(mddev));
2486                         return -EBADF;
2487                 }
2488
2489                 err = deny_bitmap_write_access(mddev->bitmap_file);
2490                 if (err) {
2491                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2492                                mdname(mddev));
2493                         fput(mddev->bitmap_file);
2494                         mddev->bitmap_file = NULL;
2495                         return err;
2496                 }
2497                 mddev->bitmap_offset = 0; /* file overrides offset */
2498         } else if (mddev->bitmap == NULL)
2499                 return -ENOENT; /* cannot remove what isn't there */
2500         err = 0;
2501         if (mddev->pers) {
2502                 mddev->pers->quiesce(mddev, 1);
2503                 if (fd >= 0)
2504                         err = bitmap_create(mddev);
2505                 if (fd < 0 || err)
2506                         bitmap_destroy(mddev);
2507                 mddev->pers->quiesce(mddev, 0);
2508         } else if (fd < 0) {
2509                 if (mddev->bitmap_file)
2510                         fput(mddev->bitmap_file);
2511                 mddev->bitmap_file = NULL;
2512         }
2513
2514         return err;
2515 }
2516
2517 /*
2518  * set_array_info is used two different ways
2519  * The original usage is when creating a new array.
2520  * In this usage, raid_disks is > 0 and it together with
2521  *  level, size, not_persistent,layout,chunksize determine the
2522  *  shape of the array.
2523  *  This will always create an array with a type-0.90.0 superblock.
2524  * The newer usage is when assembling an array.
2525  *  In this case raid_disks will be 0, and the major_version field is
2526  *  use to determine which style super-blocks are to be found on the devices.
2527  *  The minor and patch _version numbers are also kept incase the
2528  *  super_block handler wishes to interpret them.
2529  */
2530 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2531 {
2532
2533         if (info->raid_disks == 0) {
2534                 /* just setting version number for superblock loading */
2535                 if (info->major_version < 0 ||
2536                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2537                     super_types[info->major_version].name == NULL) {
2538                         /* maybe try to auto-load a module? */
2539                         printk(KERN_INFO
2540                                 "md: superblock version %d not known\n",
2541                                 info->major_version);
2542                         return -EINVAL;
2543                 }
2544                 mddev->major_version = info->major_version;
2545                 mddev->minor_version = info->minor_version;
2546                 mddev->patch_version = info->patch_version;
2547                 return 0;
2548         }
2549         mddev->major_version = MD_MAJOR_VERSION;
2550         mddev->minor_version = MD_MINOR_VERSION;
2551         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2552         mddev->ctime         = get_seconds();
2553
2554         mddev->level         = info->level;
2555         mddev->size          = info->size;
2556         mddev->raid_disks    = info->raid_disks;
2557         /* don't set md_minor, it is determined by which /dev/md* was
2558          * openned
2559          */
2560         if (info->state & (1<<MD_SB_CLEAN))
2561                 mddev->recovery_cp = MaxSector;
2562         else
2563                 mddev->recovery_cp = 0;
2564         mddev->persistent    = ! info->not_persistent;
2565
2566         mddev->layout        = info->layout;
2567         mddev->chunk_size    = info->chunk_size;
2568
2569         mddev->max_disks     = MD_SB_DISKS;
2570
2571         mddev->sb_dirty      = 1;
2572
2573         /*
2574          * Generate a 128 bit UUID
2575          */
2576         get_random_bytes(mddev->uuid, 16);
2577
2578         return 0;
2579 }
2580
2581 /*
2582  * update_array_info is used to change the configuration of an
2583  * on-line array.
2584  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2585  * fields in the info are checked against the array.
2586  * Any differences that cannot be handled will cause an error.
2587  * Normally, only one change can be managed at a time.
2588  */
2589 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2590 {
2591         int rv = 0;
2592         int cnt = 0;
2593         int state = 0;
2594
2595         /* calculate expected state,ignoring low bits */
2596         if (mddev->bitmap && mddev->bitmap_offset)
2597                 state |= (1 << MD_SB_BITMAP_PRESENT);
2598
2599         if (mddev->major_version != info->major_version ||
2600             mddev->minor_version != info->minor_version ||
2601 /*          mddev->patch_version != info->patch_version || */
2602             mddev->ctime         != info->ctime         ||
2603             mddev->level         != info->level         ||
2604 /*          mddev->layout        != info->layout        || */
2605             !mddev->persistent   != info->not_persistent||
2606             mddev->chunk_size    != info->chunk_size    ||
2607             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2608             ((state^info->state) & 0xfffffe00)
2609                 )
2610                 return -EINVAL;
2611         /* Check there is only one change */
2612         if (mddev->size != info->size) cnt++;
2613         if (mddev->raid_disks != info->raid_disks) cnt++;
2614         if (mddev->layout != info->layout) cnt++;
2615         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2616         if (cnt == 0) return 0;
2617         if (cnt > 1) return -EINVAL;
2618
2619         if (mddev->layout != info->layout) {
2620                 /* Change layout
2621                  * we don't need to do anything at the md level, the
2622                  * personality will take care of it all.
2623                  */
2624                 if (mddev->pers->reconfig == NULL)
2625                         return -EINVAL;
2626                 else
2627                         return mddev->pers->reconfig(mddev, info->layout, -1);
2628         }
2629         if (mddev->size != info->size) {
2630                 mdk_rdev_t * rdev;
2631                 struct list_head *tmp;
2632                 if (mddev->pers->resize == NULL)
2633                         return -EINVAL;
2634                 /* The "size" is the amount of each device that is used.
2635                  * This can only make sense for arrays with redundancy.
2636                  * linear and raid0 always use whatever space is available
2637                  * We can only consider changing the size if no resync
2638                  * or reconstruction is happening, and if the new size
2639                  * is acceptable. It must fit before the sb_offset or,
2640                  * if that is <data_offset, it must fit before the
2641                  * size of each device.
2642                  * If size is zero, we find the largest size that fits.
2643                  */
2644                 if (mddev->sync_thread)
2645                         return -EBUSY;
2646                 ITERATE_RDEV(mddev,rdev,tmp) {
2647                         sector_t avail;
2648                         int fit = (info->size == 0);
2649                         if (rdev->sb_offset > rdev->data_offset)
2650                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2651                         else
2652                                 avail = get_capacity(rdev->bdev->bd_disk)
2653                                         - rdev->data_offset;
2654                         if (fit && (info->size == 0 || info->size > avail/2))
2655                                 info->size = avail/2;
2656                         if (avail < ((sector_t)info->size << 1))
2657                                 return -ENOSPC;
2658                 }
2659                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
2660                 if (!rv) {
2661                         struct block_device *bdev;
2662
2663                         bdev = bdget_disk(mddev->gendisk, 0);
2664                         if (bdev) {
2665                                 down(&bdev->bd_inode->i_sem);
2666                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2667                                 up(&bdev->bd_inode->i_sem);
2668                                 bdput(bdev);
2669                         }
2670                 }
2671         }
2672         if (mddev->raid_disks    != info->raid_disks) {
2673                 /* change the number of raid disks */
2674                 if (mddev->pers->reshape == NULL)
2675                         return -EINVAL;
2676                 if (info->raid_disks <= 0 ||
2677                     info->raid_disks >= mddev->max_disks)
2678                         return -EINVAL;
2679                 if (mddev->sync_thread)
2680                         return -EBUSY;
2681                 rv = mddev->pers->reshape(mddev, info->raid_disks);
2682                 if (!rv) {
2683                         struct block_device *bdev;
2684
2685                         bdev = bdget_disk(mddev->gendisk, 0);
2686                         if (bdev) {
2687                                 down(&bdev->bd_inode->i_sem);
2688                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
2689                                 up(&bdev->bd_inode->i_sem);
2690                                 bdput(bdev);
2691                         }
2692                 }
2693         }
2694         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
2695                 if (mddev->pers->quiesce == NULL)
2696                         return -EINVAL;
2697                 if (mddev->recovery || mddev->sync_thread)
2698                         return -EBUSY;
2699                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
2700                         /* add the bitmap */
2701                         if (mddev->bitmap)
2702                                 return -EEXIST;
2703                         if (mddev->default_bitmap_offset == 0)
2704                                 return -EINVAL;
2705                         mddev->bitmap_offset = mddev->default_bitmap_offset;
2706                         mddev->pers->quiesce(mddev, 1);
2707                         rv = bitmap_create(mddev);
2708                         if (rv)
2709                                 bitmap_destroy(mddev);
2710                         mddev->pers->quiesce(mddev, 0);
2711                 } else {
2712                         /* remove the bitmap */
2713                         if (!mddev->bitmap)
2714                                 return -ENOENT;
2715                         if (mddev->bitmap->file)
2716                                 return -EINVAL;
2717                         mddev->pers->quiesce(mddev, 1);
2718                         bitmap_destroy(mddev);
2719                         mddev->pers->quiesce(mddev, 0);
2720                         mddev->bitmap_offset = 0;
2721                 }
2722         }
2723         md_update_sb(mddev);
2724         return rv;
2725 }
2726
2727 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
2728 {
2729         mdk_rdev_t *rdev;
2730
2731         if (mddev->pers == NULL)
2732                 return -ENODEV;
2733
2734         rdev = find_rdev(mddev, dev);
2735         if (!rdev)
2736                 return -ENODEV;
2737
2738         md_error(mddev, rdev);
2739         return 0;
2740 }
2741
2742 static int md_ioctl(struct inode *inode, struct file *file,
2743                         unsigned int cmd, unsigned long arg)
2744 {
2745         int err = 0;
2746         void __user *argp = (void __user *)arg;
2747         struct hd_geometry __user *loc = argp;
2748         mddev_t *mddev = NULL;
2749
2750         if (!capable(CAP_SYS_ADMIN))
2751                 return -EACCES;
2752
2753         /*
2754          * Commands dealing with the RAID driver but not any
2755          * particular array:
2756          */
2757         switch (cmd)
2758         {
2759                 case RAID_VERSION:
2760                         err = get_version(argp);
2761                         goto done;
2762
2763                 case PRINT_RAID_DEBUG:
2764                         err = 0;
2765                         md_print_devices();
2766                         goto done;
2767
2768 #ifndef MODULE
2769                 case RAID_AUTORUN:
2770                         err = 0;
2771                         autostart_arrays(arg);
2772                         goto done;
2773 #endif
2774                 default:;
2775         }
2776
2777         /*
2778          * Commands creating/starting a new array:
2779          */
2780
2781         mddev = inode->i_bdev->bd_disk->private_data;
2782
2783         if (!mddev) {
2784                 BUG();
2785                 goto abort;
2786         }
2787
2788
2789         if (cmd == START_ARRAY) {
2790                 /* START_ARRAY doesn't need to lock the array as autostart_array
2791                  * does the locking, and it could even be a different array
2792                  */
2793                 static int cnt = 3;
2794                 if (cnt > 0 ) {
2795                         printk(KERN_WARNING
2796                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
2797                                "This will not be supported beyond 2.6\n",
2798                                current->comm, current->pid);
2799                         cnt--;
2800                 }
2801                 err = autostart_array(new_decode_dev(arg));
2802                 if (err) {
2803                         printk(KERN_WARNING "md: autostart failed!\n");
2804                         goto abort;
2805                 }
2806                 goto done;
2807         }
2808
2809         err = mddev_lock(mddev);
2810         if (err) {
2811                 printk(KERN_INFO
2812                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
2813                         err, cmd);
2814                 goto abort;
2815         }
2816
2817         switch (cmd)
2818         {
2819                 case SET_ARRAY_INFO:
2820                         {
2821                                 mdu_array_info_t info;
2822                                 if (!arg)
2823                                         memset(&info, 0, sizeof(info));
2824                                 else if (copy_from_user(&info, argp, sizeof(info))) {
2825                                         err = -EFAULT;
2826                                         goto abort_unlock;
2827                                 }
2828                                 if (mddev->pers) {
2829                                         err = update_array_info(mddev, &info);
2830                                         if (err) {
2831                                                 printk(KERN_WARNING "md: couldn't update"
2832                                                        " array info. %d\n", err);
2833                                                 goto abort_unlock;
2834                                         }
2835                                         goto done_unlock;
2836                                 }
2837                                 if (!list_empty(&mddev->disks)) {
2838                                         printk(KERN_WARNING
2839                                                "md: array %s already has disks!\n",
2840                                                mdname(mddev));
2841                                         err = -EBUSY;
2842                                         goto abort_unlock;
2843                                 }
2844                                 if (mddev->raid_disks) {
2845                                         printk(KERN_WARNING
2846                                                "md: array %s already initialised!\n",
2847                                                mdname(mddev));
2848                                         err = -EBUSY;
2849                                         goto abort_unlock;
2850                                 }
2851                                 err = set_array_info(mddev, &info);
2852                                 if (err) {
2853                                         printk(KERN_WARNING "md: couldn't set"
2854                                                " array info. %d\n", err);
2855                                         goto abort_unlock;
2856                                 }
2857                         }
2858                         goto done_unlock;
2859
2860                 default:;
2861         }
2862
2863         /*
2864          * Commands querying/configuring an existing array:
2865          */
2866         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
2867          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
2868         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
2869                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
2870                 err = -ENODEV;
2871                 goto abort_unlock;
2872         }
2873
2874         /*
2875          * Commands even a read-only array can execute:
2876          */
2877         switch (cmd)
2878         {
2879                 case GET_ARRAY_INFO:
2880                         err = get_array_info(mddev, argp);
2881                         goto done_unlock;
2882
2883                 case GET_BITMAP_FILE:
2884                         err = get_bitmap_file(mddev, argp);
2885                         goto done_unlock;
2886
2887                 case GET_DISK_INFO:
2888                         err = get_disk_info(mddev, argp);
2889                         goto done_unlock;
2890
2891                 case RESTART_ARRAY_RW:
2892                         err = restart_array(mddev);
2893                         goto done_unlock;
2894
2895                 case STOP_ARRAY:
2896                         err = do_md_stop (mddev, 0);
2897                         goto done_unlock;
2898
2899                 case STOP_ARRAY_RO:
2900                         err = do_md_stop (mddev, 1);
2901                         goto done_unlock;
2902
2903         /*
2904          * We have a problem here : there is no easy way to give a CHS
2905          * virtual geometry. We currently pretend that we have a 2 heads
2906          * 4 sectors (with a BIG number of cylinders...). This drives
2907          * dosfs just mad... ;-)
2908          */
2909                 case HDIO_GETGEO:
2910                         if (!loc) {
2911                                 err = -EINVAL;
2912                                 goto abort_unlock;
2913                         }
2914                         err = put_user (2, (char __user *) &loc->heads);
2915                         if (err)
2916                                 goto abort_unlock;
2917                         err = put_user (4, (char __user *) &loc->sectors);
2918                         if (err)
2919                                 goto abort_unlock;
2920                         err = put_user(get_capacity(mddev->gendisk)/8,
2921                                         (short __user *) &loc->cylinders);
2922                         if (err)
2923                                 goto abort_unlock;
2924                         err = put_user (get_start_sect(inode->i_bdev),
2925                                                 (long __user *) &loc->start);
2926                         goto done_unlock;
2927         }
2928
2929         /*
2930          * The remaining ioctls are changing the state of the
2931          * superblock, so we do not allow read-only arrays
2932          * here:
2933          */
2934         if (mddev->ro) {
2935                 err = -EROFS;
2936                 goto abort_unlock;
2937         }
2938
2939         switch (cmd)
2940         {
2941                 case ADD_NEW_DISK:
2942                 {
2943                         mdu_disk_info_t info;
2944                         if (copy_from_user(&info, argp, sizeof(info)))
2945                                 err = -EFAULT;
2946                         else
2947                                 err = add_new_disk(mddev, &info);
2948                         goto done_unlock;
2949                 }
2950
2951                 case HOT_REMOVE_DISK:
2952                         err = hot_remove_disk(mddev, new_decode_dev(arg));
2953                         goto done_unlock;
2954
2955                 case HOT_ADD_DISK:
2956                         err = hot_add_disk(mddev, new_decode_dev(arg));
2957                         goto done_unlock;
2958
2959                 case SET_DISK_FAULTY:
2960                         err = set_disk_faulty(mddev, new_decode_dev(arg));
2961                         goto done_unlock;
2962
2963                 case RUN_ARRAY:
2964                         err = do_md_run (mddev);
2965                         goto done_unlock;
2966
2967                 case SET_BITMAP_FILE:
2968                         err = set_bitmap_file(mddev, (int)arg);
2969                         goto done_unlock;
2970
2971                 default:
2972                         if (_IOC_TYPE(cmd) == MD_MAJOR)
2973                                 printk(KERN_WARNING "md: %s(pid %d) used"
2974                                         " obsolete MD ioctl, upgrade your"
2975                                         " software to use new ictls.\n",
2976                                         current->comm, current->pid);
2977                         err = -EINVAL;
2978                         goto abort_unlock;
2979         }
2980
2981 done_unlock:
2982 abort_unlock:
2983         mddev_unlock(mddev);
2984
2985         return err;
2986 done:
2987         if (err)
2988                 MD_BUG();
2989 abort:
2990         return err;
2991 }
2992
2993 static int md_open(struct inode *inode, struct file *file)
2994 {
2995         /*
2996          * Succeed if we can lock the mddev, which confirms that
2997          * it isn't being stopped right now.
2998          */
2999         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3000         int err;
3001
3002         if ((err = mddev_lock(mddev)))
3003                 goto out;
3004
3005         err = 0;
3006         mddev_get(mddev);
3007         mddev_unlock(mddev);
3008
3009         check_disk_change(inode->i_bdev);
3010  out:
3011         return err;
3012 }
3013
3014 static int md_release(struct inode *inode, struct file * file)
3015 {
3016         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3017
3018         if (!mddev)
3019                 BUG();
3020         mddev_put(mddev);
3021
3022         return 0;
3023 }
3024
3025 static int md_media_changed(struct gendisk *disk)
3026 {
3027         mddev_t *mddev = disk->private_data;
3028
3029         return mddev->changed;
3030 }
3031
3032 static int md_revalidate(struct gendisk *disk)
3033 {
3034         mddev_t *mddev = disk->private_data;
3035
3036         mddev->changed = 0;
3037         return 0;
3038 }
3039 static struct block_device_operations md_fops =
3040 {
3041         .owner          = THIS_MODULE,
3042         .open           = md_open,
3043         .release        = md_release,
3044         .ioctl          = md_ioctl,
3045         .media_changed  = md_media_changed,
3046         .revalidate_disk= md_revalidate,
3047 };
3048
3049 static int md_thread(void * arg)
3050 {
3051         mdk_thread_t *thread = arg;
3052
3053         /*
3054          * md_thread is a 'system-thread', it's priority should be very
3055          * high. We avoid resource deadlocks individually in each
3056          * raid personality. (RAID5 does preallocation) We also use RR and
3057          * the very same RT priority as kswapd, thus we will never get
3058          * into a priority inversion deadlock.
3059          *
3060          * we definitely have to have equal or higher priority than
3061          * bdflush, otherwise bdflush will deadlock if there are too
3062          * many dirty RAID5 blocks.
3063          */
3064
3065         complete(thread->event);
3066         while (!kthread_should_stop()) {
3067                 void (*run)(mddev_t *);
3068
3069                 wait_event_interruptible_timeout(thread->wqueue,
3070                                                  test_bit(THREAD_WAKEUP, &thread->flags)
3071                                                  || kthread_should_stop(),
3072                                                  thread->timeout);
3073                 try_to_freeze();
3074
3075                 clear_bit(THREAD_WAKEUP, &thread->flags);
3076
3077                 run = thread->run;
3078                 if (run)
3079                         run(thread->mddev);
3080         }
3081
3082         return 0;
3083 }
3084
3085 void md_wakeup_thread(mdk_thread_t *thread)
3086 {
3087         if (thread) {
3088                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3089                 set_bit(THREAD_WAKEUP, &thread->flags);
3090                 wake_up(&thread->wqueue);
3091         }
3092 }
3093
3094 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3095                                  const char *name)
3096 {
3097         mdk_thread_t *thread;
3098         struct completion event;
3099
3100         thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3101         if (!thread)
3102                 return NULL;
3103
3104         memset(thread, 0, sizeof(mdk_thread_t));
3105         init_waitqueue_head(&thread->wqueue);
3106
3107         init_completion(&event);
3108         thread->event = &event;
3109         thread->run = run;
3110         thread->mddev = mddev;
3111         thread->name = name;
3112         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3113         thread->tsk = kthread_run(md_thread, thread, mdname(thread->mddev));
3114         if (IS_ERR(thread->tsk)) {
3115                 kfree(thread);
3116                 return NULL;
3117         }
3118         wait_for_completion(&event);
3119         return thread;
3120 }
3121
3122 void md_unregister_thread(mdk_thread_t *thread)
3123 {
3124         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3125
3126         kthread_stop(thread->tsk);
3127         kfree(thread);
3128 }
3129
3130 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3131 {
3132         if (!mddev) {
3133                 MD_BUG();
3134                 return;
3135         }
3136
3137         if (!rdev || rdev->faulty)
3138                 return;
3139 /*
3140         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3141                 mdname(mddev),
3142                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3143                 __builtin_return_address(0),__builtin_return_address(1),
3144                 __builtin_return_address(2),__builtin_return_address(3));
3145 */
3146         if (!mddev->pers->error_handler)
3147                 return;
3148         mddev->pers->error_handler(mddev,rdev);
3149         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3150         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3151         md_wakeup_thread(mddev->thread);
3152 }
3153
3154 /* seq_file implementation /proc/mdstat */
3155
3156 static void status_unused(struct seq_file *seq)
3157 {
3158         int i = 0;
3159         mdk_rdev_t *rdev;
3160         struct list_head *tmp;
3161
3162         seq_printf(seq, "unused devices: ");
3163
3164         ITERATE_RDEV_PENDING(rdev,tmp) {
3165                 char b[BDEVNAME_SIZE];
3166                 i++;
3167                 seq_printf(seq, "%s ",
3168                               bdevname(rdev->bdev,b));
3169         }
3170         if (!i)
3171                 seq_printf(seq, "<none>");
3172
3173         seq_printf(seq, "\n");
3174 }
3175
3176
3177 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3178 {
3179         unsigned long max_blocks, resync, res, dt, db, rt;
3180
3181         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3182
3183         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3184                 max_blocks = mddev->resync_max_sectors >> 1;
3185         else
3186                 max_blocks = mddev->size;
3187
3188         /*
3189          * Should not happen.
3190          */
3191         if (!max_blocks) {
3192                 MD_BUG();
3193                 return;
3194         }
3195         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3196         {
3197                 int i, x = res/50, y = 20-x;
3198                 seq_printf(seq, "[");
3199                 for (i = 0; i < x; i++)
3200                         seq_printf(seq, "=");
3201                 seq_printf(seq, ">");
3202                 for (i = 0; i < y; i++)
3203                         seq_printf(seq, ".");
3204                 seq_printf(seq, "] ");
3205         }
3206         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3207                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3208                        "resync" : "recovery"),
3209                       res/10, res % 10, resync, max_blocks);
3210
3211         /*
3212          * We do not want to overflow, so the order of operands and
3213          * the * 100 / 100 trick are important. We do a +1 to be
3214          * safe against division by zero. We only estimate anyway.
3215          *
3216          * dt: time from mark until now
3217          * db: blocks written from mark until now
3218          * rt: remaining time
3219          */
3220         dt = ((jiffies - mddev->resync_mark) / HZ);
3221         if (!dt) dt++;
3222         db = resync - (mddev->resync_mark_cnt/2);
3223         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3224
3225         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3226
3227         seq_printf(seq, " speed=%ldK/sec", db/dt);
3228 }
3229
3230 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3231 {
3232         struct list_head *tmp;
3233         loff_t l = *pos;
3234         mddev_t *mddev;
3235
3236         if (l >= 0x10000)
3237                 return NULL;
3238         if (!l--)
3239                 /* header */
3240                 return (void*)1;
3241
3242         spin_lock(&all_mddevs_lock);
3243         list_for_each(tmp,&all_mddevs)
3244                 if (!l--) {
3245                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3246                         mddev_get(mddev);
3247                         spin_unlock(&all_mddevs_lock);
3248                         return mddev;
3249                 }
3250         spin_unlock(&all_mddevs_lock);
3251         if (!l--)
3252                 return (void*)2;/* tail */
3253         return NULL;
3254 }
3255
3256 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3257 {
3258         struct list_head *tmp;
3259         mddev_t *next_mddev, *mddev = v;
3260
3261         ++*pos;
3262         if (v == (void*)2)
3263                 return NULL;
3264
3265         spin_lock(&all_mddevs_lock);
3266         if (v == (void*)1)
3267                 tmp = all_mddevs.next;
3268         else
3269                 tmp = mddev->all_mddevs.next;
3270         if (tmp != &all_mddevs)
3271                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3272         else {
3273                 next_mddev = (void*)2;
3274                 *pos = 0x10000;
3275         }
3276         spin_unlock(&all_mddevs_lock);
3277
3278         if (v != (void*)1)
3279                 mddev_put(mddev);
3280         return next_mddev;
3281
3282 }
3283
3284 static void md_seq_stop(struct seq_file *seq, void *v)
3285 {
3286         mddev_t *mddev = v;
3287
3288         if (mddev && v != (void*)1 && v != (void*)2)
3289                 mddev_put(mddev);
3290 }
3291
3292 static int md_seq_show(struct seq_file *seq, void *v)
3293 {
3294         mddev_t *mddev = v;
3295         sector_t size;
3296         struct list_head *tmp2;
3297         mdk_rdev_t *rdev;
3298         int i;
3299         struct bitmap *bitmap;
3300
3301         if (v == (void*)1) {
3302                 seq_printf(seq, "Personalities : ");
3303                 spin_lock(&pers_lock);
3304                 for (i = 0; i < MAX_PERSONALITY; i++)
3305                         if (pers[i])
3306                                 seq_printf(seq, "[%s] ", pers[i]->name);
3307
3308                 spin_unlock(&pers_lock);
3309                 seq_printf(seq, "\n");
3310                 return 0;
3311         }
3312         if (v == (void*)2) {
3313                 status_unused(seq);
3314                 return 0;
3315         }
3316
3317         if (mddev_lock(mddev)!=0)
3318                 return -EINTR;
3319         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3320                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3321                                                 mddev->pers ? "" : "in");
3322                 if (mddev->pers) {
3323                         if (mddev->ro)
3324                                 seq_printf(seq, " (read-only)");
3325                         seq_printf(seq, " %s", mddev->pers->name);
3326                 }
3327
3328                 size = 0;
3329                 ITERATE_RDEV(mddev,rdev,tmp2) {
3330                         char b[BDEVNAME_SIZE];
3331                         seq_printf(seq, " %s[%d]",
3332                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3333                         if (test_bit(WriteMostly, &rdev->flags))
3334                                 seq_printf(seq, "(W)");
3335                         if (rdev->faulty) {
3336                                 seq_printf(seq, "(F)");
3337                                 continue;
3338                         }
3339                         size += rdev->size;
3340                 }
3341
3342                 if (!list_empty(&mddev->disks)) {
3343                         if (mddev->pers)
3344                                 seq_printf(seq, "\n      %llu blocks",
3345                                         (unsigned long long)mddev->array_size);
3346                         else
3347                                 seq_printf(seq, "\n      %llu blocks",
3348                                         (unsigned long long)size);
3349                 }
3350
3351                 if (mddev->pers) {
3352                         mddev->pers->status (seq, mddev);
3353                         seq_printf(seq, "\n      ");
3354                         if (mddev->curr_resync > 2) {
3355                                 status_resync (seq, mddev);
3356                                 seq_printf(seq, "\n      ");
3357                         } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3358                                 seq_printf(seq, "       resync=DELAYED\n      ");
3359                 } else
3360                         seq_printf(seq, "\n       ");
3361
3362                 if ((bitmap = mddev->bitmap)) {
3363                         unsigned long chunk_kb;
3364                         unsigned long flags;
3365                         spin_lock_irqsave(&bitmap->lock, flags);
3366                         chunk_kb = bitmap->chunksize >> 10;
3367                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3368                                 "%lu%s chunk",
3369                                 bitmap->pages - bitmap->missing_pages,
3370                                 bitmap->pages,
3371                                 (bitmap->pages - bitmap->missing_pages)
3372                                         << (PAGE_SHIFT - 10),
3373                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3374                                 chunk_kb ? "KB" : "B");
3375                         if (bitmap->file) {
3376                                 seq_printf(seq, ", file: ");
3377                                 seq_path(seq, bitmap->file->f_vfsmnt,
3378                                          bitmap->file->f_dentry," \t\n");
3379                         }
3380
3381                         seq_printf(seq, "\n");
3382                         spin_unlock_irqrestore(&bitmap->lock, flags);
3383                 }
3384
3385                 seq_printf(seq, "\n");
3386         }
3387         mddev_unlock(mddev);
3388
3389         return 0;
3390 }
3391
3392 static struct seq_operations md_seq_ops = {
3393         .start  = md_seq_start,
3394         .next   = md_seq_next,
3395         .stop   = md_seq_stop,
3396         .show   = md_seq_show,
3397 };
3398
3399 static int md_seq_open(struct inode *inode, struct file *file)
3400 {
3401         int error;
3402
3403         error = seq_open(file, &md_seq_ops);
3404         return error;
3405 }
3406
3407 static struct file_operations md_seq_fops = {
3408         .open           = md_seq_open,
3409         .read           = seq_read,
3410         .llseek         = seq_lseek,
3411         .release        = seq_release,
3412 };
3413
3414 int register_md_personality(int pnum, mdk_personality_t *p)
3415 {
3416         if (pnum >= MAX_PERSONALITY) {
3417                 printk(KERN_ERR
3418                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3419                        p->name, pnum, MAX_PERSONALITY-1);
3420                 return -EINVAL;
3421         }
3422
3423         spin_lock(&pers_lock);
3424         if (pers[pnum]) {
3425                 spin_unlock(&pers_lock);
3426                 return -EBUSY;
3427         }
3428
3429         pers[pnum] = p;
3430         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3431         spin_unlock(&pers_lock);
3432         return 0;
3433 }
3434
3435 int unregister_md_personality(int pnum)
3436 {
3437         if (pnum >= MAX_PERSONALITY)
3438                 return -EINVAL;
3439
3440         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3441         spin_lock(&pers_lock);
3442         pers[pnum] = NULL;
3443         spin_unlock(&pers_lock);
3444         return 0;
3445 }
3446
3447 static int is_mddev_idle(mddev_t *mddev)
3448 {
3449         mdk_rdev_t * rdev;
3450         struct list_head *tmp;
3451         int idle;
3452         unsigned long curr_events;
3453
3454         idle = 1;
3455         ITERATE_RDEV(mddev,rdev,tmp) {
3456                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3457                 curr_events = disk_stat_read(disk, read_sectors) +
3458                                 disk_stat_read(disk, write_sectors) -
3459                                 atomic_read(&disk->sync_io);
3460                 /* Allow some slack between valud of curr_events and last_events,
3461                  * as there are some uninteresting races.
3462                  * Note: the following is an unsigned comparison.
3463                  */
3464                 if ((curr_events - rdev->last_events + 32) > 64) {
3465                         rdev->last_events = curr_events;
3466                         idle = 0;
3467                 }
3468         }
3469         return idle;
3470 }
3471
3472 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3473 {
3474         /* another "blocks" (512byte) blocks have been synced */
3475         atomic_sub(blocks, &mddev->recovery_active);
3476         wake_up(&mddev->recovery_wait);
3477         if (!ok) {
3478                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3479                 md_wakeup_thread(mddev->thread);
3480                 // stop recovery, signal do_sync ....
3481         }
3482 }
3483
3484
3485 /* md_write_start(mddev, bi)
3486  * If we need to update some array metadata (e.g. 'active' flag
3487  * in superblock) before writing, schedule a superblock update
3488  * and wait for it to complete.
3489  */
3490 void md_write_start(mddev_t *mddev, struct bio *bi)
3491 {
3492         if (bio_data_dir(bi) != WRITE)
3493                 return;
3494
3495         atomic_inc(&mddev->writes_pending);
3496         if (mddev->in_sync) {
3497                 spin_lock(&mddev->write_lock);
3498                 if (mddev->in_sync) {
3499                         mddev->in_sync = 0;
3500                         mddev->sb_dirty = 1;
3501                         md_wakeup_thread(mddev->thread);
3502                 }
3503                 spin_unlock(&mddev->write_lock);
3504         }
3505         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3506 }
3507
3508 void md_write_end(mddev_t *mddev)
3509 {
3510         if (atomic_dec_and_test(&mddev->writes_pending)) {
3511                 if (mddev->safemode == 2)
3512                         md_wakeup_thread(mddev->thread);
3513                 else
3514                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3515         }
3516 }
3517
3518 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3519
3520 #define SYNC_MARKS      10
3521 #define SYNC_MARK_STEP  (3*HZ)
3522 static void md_do_sync(mddev_t *mddev)
3523 {
3524         mddev_t *mddev2;
3525         unsigned int currspeed = 0,
3526                  window;
3527         sector_t max_sectors,j, io_sectors;
3528         unsigned long mark[SYNC_MARKS];
3529         sector_t mark_cnt[SYNC_MARKS];
3530         int last_mark,m;
3531         struct list_head *tmp;
3532         sector_t last_check;
3533         int skipped = 0;
3534
3535         /* just incase thread restarts... */
3536         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3537                 return;
3538
3539         /* we overload curr_resync somewhat here.
3540          * 0 == not engaged in resync at all
3541          * 2 == checking that there is no conflict with another sync
3542          * 1 == like 2, but have yielded to allow conflicting resync to
3543          *              commense
3544          * other == active in resync - this many blocks
3545          *
3546          * Before starting a resync we must have set curr_resync to
3547          * 2, and then checked that every "conflicting" array has curr_resync
3548          * less than ours.  When we find one that is the same or higher
3549          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3550          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3551          * This will mean we have to start checking from the beginning again.
3552          *
3553          */
3554
3555         do {
3556                 mddev->curr_resync = 2;
3557
3558         try_again:
3559                 if (signal_pending(current)) {
3560                         flush_signals(current);
3561                         goto skip;
3562                 }
3563                 ITERATE_MDDEV(mddev2,tmp) {
3564                         if (mddev2 == mddev)
3565                                 continue;
3566                         if (mddev2->curr_resync &&
3567                             match_mddev_units(mddev,mddev2)) {
3568                                 DEFINE_WAIT(wq);
3569                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3570                                         /* arbitrarily yield */
3571                                         mddev->curr_resync = 1;
3572                                         wake_up(&resync_wait);
3573                                 }
3574                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3575                                         /* no need to wait here, we can wait the next
3576                                          * time 'round when curr_resync == 2
3577                                          */
3578                                         continue;
3579                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3580                                 if (!signal_pending(current)
3581                                     && mddev2->curr_resync >= mddev->curr_resync) {
3582                                         printk(KERN_INFO "md: delaying resync of %s"
3583                                                " until %s has finished resync (they"
3584                                                " share one or more physical units)\n",
3585                                                mdname(mddev), mdname(mddev2));
3586                                         mddev_put(mddev2);
3587                                         schedule();
3588                                         finish_wait(&resync_wait, &wq);
3589                                         goto try_again;
3590                                 }
3591                                 finish_wait(&resync_wait, &wq);
3592                         }
3593                 }
3594         } while (mddev->curr_resync < 2);
3595
3596         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3597                 /* resync follows the size requested by the personality,
3598                  * which defaults to physical size, but can be virtual size
3599                  */
3600                 max_sectors = mddev->resync_max_sectors;
3601         else
3602                 /* recovery follows the physical size of devices */
3603                 max_sectors = mddev->size << 1;
3604
3605         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3606         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3607                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3608         printk(KERN_INFO "md: using maximum available idle IO bandwith "
3609                "(but not more than %d KB/sec) for reconstruction.\n",
3610                sysctl_speed_limit_max);
3611
3612         is_mddev_idle(mddev); /* this also initializes IO event counters */
3613         /* we don't use the checkpoint if there's a bitmap */
3614         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap)
3615                 j = mddev->recovery_cp;
3616         else
3617                 j = 0;
3618         io_sectors = 0;
3619         for (m = 0; m < SYNC_MARKS; m++) {
3620                 mark[m] = jiffies;
3621                 mark_cnt[m] = io_sectors;
3622         }
3623         last_mark = 0;
3624         mddev->resync_mark = mark[last_mark];
3625         mddev->resync_mark_cnt = mark_cnt[last_mark];
3626
3627         /*
3628          * Tune reconstruction:
3629          */
3630         window = 32*(PAGE_SIZE/512);
3631         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
3632                 window/2,(unsigned long long) max_sectors/2);
3633
3634         atomic_set(&mddev->recovery_active, 0);
3635         init_waitqueue_head(&mddev->recovery_wait);
3636         last_check = 0;
3637
3638         if (j>2) {
3639                 printk(KERN_INFO
3640                         "md: resuming recovery of %s from checkpoint.\n",
3641                         mdname(mddev));
3642                 mddev->curr_resync = j;
3643         }
3644
3645         while (j < max_sectors) {
3646                 sector_t sectors;
3647
3648                 skipped = 0;
3649                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
3650                                             currspeed < sysctl_speed_limit_min);
3651                 if (sectors == 0) {
3652                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3653                         goto out;
3654                 }
3655
3656                 if (!skipped) { /* actual IO requested */
3657                         io_sectors += sectors;
3658                         atomic_add(sectors, &mddev->recovery_active);
3659                 }
3660
3661                 j += sectors;
3662                 if (j>1) mddev->curr_resync = j;
3663
3664
3665                 if (last_check + window > io_sectors || j == max_sectors)
3666                         continue;
3667
3668                 last_check = io_sectors;
3669
3670                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
3671                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
3672                         break;
3673
3674         repeat:
3675                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
3676                         /* step marks */
3677                         int next = (last_mark+1) % SYNC_MARKS;
3678
3679                         mddev->resync_mark = mark[next];
3680                         mddev->resync_mark_cnt = mark_cnt[next];
3681                         mark[next] = jiffies;
3682                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
3683                         last_mark = next;
3684                 }
3685
3686
3687                 if (signal_pending(current)) {
3688                         /*
3689                          * got a signal, exit.
3690                          */
3691                         printk(KERN_INFO
3692                                 "md: md_do_sync() got signal ... exiting\n");
3693                         flush_signals(current);
3694                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3695                         goto out;
3696                 }
3697
3698                 /*
3699                  * this loop exits only if either when we are slower than
3700                  * the 'hard' speed limit, or the system was IO-idle for
3701                  * a jiffy.
3702                  * the system might be non-idle CPU-wise, but we only care
3703                  * about not overloading the IO subsystem. (things like an
3704                  * e2fsck being done on the RAID array should execute fast)
3705                  */
3706                 mddev->queue->unplug_fn(mddev->queue);
3707                 cond_resched();
3708
3709                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
3710                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
3711
3712                 if (currspeed > sysctl_speed_limit_min) {
3713                         if ((currspeed > sysctl_speed_limit_max) ||
3714                                         !is_mddev_idle(mddev)) {
3715                                 msleep_interruptible(250);
3716                                 goto repeat;
3717                         }
3718                 }
3719         }
3720         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
3721         /*
3722          * this also signals 'finished resyncing' to md_stop
3723          */
3724  out:
3725         mddev->queue->unplug_fn(mddev->queue);
3726
3727         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
3728
3729         /* tell personality that we are finished */
3730         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
3731
3732         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3733             mddev->curr_resync > 2 &&
3734             mddev->curr_resync >= mddev->recovery_cp) {
3735                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3736                         printk(KERN_INFO
3737                                 "md: checkpointing recovery of %s.\n",
3738                                 mdname(mddev));
3739                         mddev->recovery_cp = mddev->curr_resync;
3740                 } else
3741                         mddev->recovery_cp = MaxSector;
3742         }
3743
3744  skip:
3745         mddev->curr_resync = 0;
3746         wake_up(&resync_wait);
3747         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
3748         md_wakeup_thread(mddev->thread);
3749 }
3750
3751
3752 /*
3753  * This routine is regularly called by all per-raid-array threads to
3754  * deal with generic issues like resync and super-block update.
3755  * Raid personalities that don't have a thread (linear/raid0) do not
3756  * need this as they never do any recovery or update the superblock.
3757  *
3758  * It does not do any resync itself, but rather "forks" off other threads
3759  * to do that as needed.
3760  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
3761  * "->recovery" and create a thread at ->sync_thread.
3762  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
3763  * and wakeups up this thread which will reap the thread and finish up.
3764  * This thread also removes any faulty devices (with nr_pending == 0).
3765  *
3766  * The overall approach is:
3767  *  1/ if the superblock needs updating, update it.
3768  *  2/ If a recovery thread is running, don't do anything else.
3769  *  3/ If recovery has finished, clean up, possibly marking spares active.
3770  *  4/ If there are any faulty devices, remove them.
3771  *  5/ If array is degraded, try to add spares devices
3772  *  6/ If array has spares or is not in-sync, start a resync thread.
3773  */
3774 void md_check_recovery(mddev_t *mddev)
3775 {
3776         mdk_rdev_t *rdev;
3777         struct list_head *rtmp;
3778
3779
3780         if (mddev->bitmap)
3781                 bitmap_daemon_work(mddev->bitmap);
3782
3783         if (mddev->ro)
3784                 return;
3785
3786         if (signal_pending(current)) {
3787                 if (mddev->pers->sync_request) {
3788                         printk(KERN_INFO "md: %s in immediate safe mode\n",
3789                                mdname(mddev));
3790                         mddev->safemode = 2;
3791                 }
3792                 flush_signals(current);
3793         }
3794
3795         if ( ! (
3796                 mddev->sb_dirty ||
3797                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
3798                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
3799                 (mddev->safemode == 1) ||
3800                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
3801                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
3802                 ))
3803                 return;
3804
3805         if (mddev_trylock(mddev)==0) {
3806                 int spares =0;
3807
3808                 spin_lock(&mddev->write_lock);
3809                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
3810                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
3811                         mddev->in_sync = 1;
3812                         mddev->sb_dirty = 1;
3813                 }
3814                 if (mddev->safemode == 1)
3815                         mddev->safemode = 0;
3816                 spin_unlock(&mddev->write_lock);
3817
3818                 if (mddev->sb_dirty)
3819                         md_update_sb(mddev);
3820
3821
3822                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
3823                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
3824                         /* resync/recovery still happening */
3825                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3826                         goto unlock;
3827                 }
3828                 if (mddev->sync_thread) {
3829                         /* resync has finished, collect result */
3830                         md_unregister_thread(mddev->sync_thread);
3831                         mddev->sync_thread = NULL;
3832                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
3833                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
3834                                 /* success...*/
3835                                 /* activate any spares */
3836                                 mddev->pers->spare_active(mddev);
3837                         }
3838                         md_update_sb(mddev);
3839
3840                         /* if array is no-longer degraded, then any saved_raid_disk
3841                          * information must be scrapped
3842                          */
3843                         if (!mddev->degraded)
3844                                 ITERATE_RDEV(mddev,rdev,rtmp)
3845                                         rdev->saved_raid_disk = -1;
3846
3847                         mddev->recovery = 0;
3848                         /* flag recovery needed just to double check */
3849                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3850                         goto unlock;
3851                 }
3852                 if (mddev->recovery)
3853                         /* probably just the RECOVERY_NEEDED flag */
3854                         mddev->recovery = 0;
3855
3856                 /* no recovery is running.
3857                  * remove any failed drives, then
3858                  * add spares if possible.
3859                  * Spare are also removed and re-added, to allow
3860                  * the personality to fail the re-add.
3861                  */
3862                 ITERATE_RDEV(mddev,rdev,rtmp)
3863                         if (rdev->raid_disk >= 0 &&
3864                             (rdev->faulty || ! rdev->in_sync) &&
3865                             atomic_read(&rdev->nr_pending)==0) {
3866                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0)
3867                                         rdev->raid_disk = -1;
3868                         }
3869
3870                 if (mddev->degraded) {
3871                         ITERATE_RDEV(mddev,rdev,rtmp)
3872                                 if (rdev->raid_disk < 0
3873                                     && !rdev->faulty) {
3874                                         if (mddev->pers->hot_add_disk(mddev,rdev))
3875                                                 spares++;
3876                                         else
3877                                                 break;
3878                                 }
3879                 }
3880
3881                 if (!spares && (mddev->recovery_cp == MaxSector )) {
3882                         /* nothing we can do ... */
3883                         goto unlock;
3884                 }
3885                 if (mddev->pers->sync_request) {
3886                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3887                         if (!spares)
3888                                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3889                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
3890                                 /* We are adding a device or devices to an array
3891                                  * which has the bitmap stored on all devices.
3892                                  * So make sure all bitmap pages get written
3893                                  */
3894                                 bitmap_write_all(mddev->bitmap);
3895                         }
3896                         mddev->sync_thread = md_register_thread(md_do_sync,
3897                                                                 mddev,
3898                                                                 "%s_resync");
3899                         if (!mddev->sync_thread) {
3900                                 printk(KERN_ERR "%s: could not start resync"
3901                                         " thread...\n",
3902                                         mdname(mddev));
3903                                 /* leave the spares where they are, it shouldn't hurt */
3904                                 mddev->recovery = 0;
3905                         } else {
3906                                 md_wakeup_thread(mddev->sync_thread);
3907                         }
3908                 }
3909         unlock:
3910                 mddev_unlock(mddev);
3911         }
3912 }
3913
3914 static int md_notify_reboot(struct notifier_block *this,
3915                             unsigned long code, void *x)
3916 {
3917         struct list_head *tmp;
3918         mddev_t *mddev;
3919
3920         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
3921
3922                 printk(KERN_INFO "md: stopping all md devices.\n");
3923
3924                 ITERATE_MDDEV(mddev,tmp)
3925                         if (mddev_trylock(mddev)==0)
3926                                 do_md_stop (mddev, 1);
3927                 /*
3928                  * certain more exotic SCSI devices are known to be
3929                  * volatile wrt too early system reboots. While the
3930                  * right place to handle this issue is the given
3931                  * driver, we do want to have a safe RAID driver ...
3932                  */
3933                 mdelay(1000*1);
3934         }
3935         return NOTIFY_DONE;
3936 }
3937
3938 static struct notifier_block md_notifier = {
3939         .notifier_call  = md_notify_reboot,
3940         .next           = NULL,
3941         .priority       = INT_MAX, /* before any real devices */
3942 };
3943
3944 static void md_geninit(void)
3945 {
3946         struct proc_dir_entry *p;
3947
3948         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
3949
3950         p = create_proc_entry("mdstat", S_IRUGO, NULL);
3951         if (p)
3952                 p->proc_fops = &md_seq_fops;
3953 }
3954
3955 static int __init md_init(void)
3956 {
3957         int minor;
3958
3959         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
3960                         " MD_SB_DISKS=%d\n",
3961                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
3962                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
3963         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR,
3964                         BITMAP_MINOR);
3965
3966         if (register_blkdev(MAJOR_NR, "md"))
3967                 return -1;
3968         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
3969                 unregister_blkdev(MAJOR_NR, "md");
3970                 return -1;
3971         }
3972         devfs_mk_dir("md");
3973         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
3974                                 md_probe, NULL, NULL);
3975         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
3976                             md_probe, NULL, NULL);
3977
3978         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3979                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
3980                                 S_IFBLK|S_IRUSR|S_IWUSR,
3981                                 "md/%d", minor);
3982
3983         for (minor=0; minor < MAX_MD_DEVS; ++minor)
3984                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
3985                               S_IFBLK|S_IRUSR|S_IWUSR,
3986                               "md/mdp%d", minor);
3987
3988
3989         register_reboot_notifier(&md_notifier);
3990         raid_table_header = register_sysctl_table(raid_root_table, 1);
3991
3992         md_geninit();
3993         return (0);
3994 }
3995
3996
3997 #ifndef MODULE
3998
3999 /*
4000  * Searches all registered partitions for autorun RAID arrays
4001  * at boot time.
4002  */
4003 static dev_t detected_devices[128];
4004 static int dev_cnt;
4005
4006 void md_autodetect_dev(dev_t dev)
4007 {
4008         if (dev_cnt >= 0 && dev_cnt < 127)
4009                 detected_devices[dev_cnt++] = dev;
4010 }
4011
4012
4013 static void autostart_arrays(int part)
4014 {
4015         mdk_rdev_t *rdev;
4016         int i;
4017
4018         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4019
4020         for (i = 0; i < dev_cnt; i++) {
4021                 dev_t dev = detected_devices[i];
4022
4023                 rdev = md_import_device(dev,0, 0);
4024                 if (IS_ERR(rdev))
4025                         continue;
4026
4027                 if (rdev->faulty) {
4028                         MD_BUG();
4029                         continue;
4030                 }
4031                 list_add(&rdev->same_set, &pending_raid_disks);
4032         }
4033         dev_cnt = 0;
4034
4035         autorun_devices(part);
4036 }
4037
4038 #endif
4039
4040 static __exit void md_exit(void)
4041 {
4042         mddev_t *mddev;
4043         struct list_head *tmp;
4044         int i;
4045         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4046         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4047         for (i=0; i < MAX_MD_DEVS; i++)
4048                 devfs_remove("md/%d", i);
4049         for (i=0; i < MAX_MD_DEVS; i++)
4050                 devfs_remove("md/d%d", i);
4051
4052         devfs_remove("md");
4053
4054         unregister_blkdev(MAJOR_NR,"md");
4055         unregister_blkdev(mdp_major, "mdp");
4056         unregister_reboot_notifier(&md_notifier);
4057         unregister_sysctl_table(raid_table_header);
4058         remove_proc_entry("mdstat", NULL);
4059         ITERATE_MDDEV(mddev,tmp) {
4060                 struct gendisk *disk = mddev->gendisk;
4061                 if (!disk)
4062                         continue;
4063                 export_array(mddev);
4064                 del_gendisk(disk);
4065                 put_disk(disk);
4066                 mddev->gendisk = NULL;
4067                 mddev_put(mddev);
4068         }
4069 }
4070
4071 module_init(md_init)
4072 module_exit(md_exit)
4073
4074 EXPORT_SYMBOL(register_md_personality);
4075 EXPORT_SYMBOL(unregister_md_personality);
4076 EXPORT_SYMBOL(md_error);
4077 EXPORT_SYMBOL(md_done_sync);
4078 EXPORT_SYMBOL(md_write_start);
4079 EXPORT_SYMBOL(md_write_end);
4080 EXPORT_SYMBOL(md_register_thread);
4081 EXPORT_SYMBOL(md_unregister_thread);
4082 EXPORT_SYMBOL(md_wakeup_thread);
4083 EXPORT_SYMBOL(md_print_devices);
4084 EXPORT_SYMBOL(md_check_recovery);
4085 MODULE_LICENSE("GPL");
4086 MODULE_ALIAS("md");
4087 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);